The interconnected school context: Meta-analyses of the associations between peer aggression involvement and teacher-student relationship closeness

Teacher-Student relationships

The Bioecological Model of development posits that development occurs within a set of interconnected social-cultural environments that reciprocally interact with the child (Bronfenbrenner & Morris, 2006). It has commonly been used to explain the adverse psychosocial consequences on children that are associated with involvement in peer aggression as well as how social contexts, including, for example, relationships that form between students and their teachers, can reduce the incidence and the impacts of peer aggression involvement (Bouchard & Smith, 2017; Wang et al., 2015). The most proximate of the model systems is the microsystem, which comprises the direct interactions that children have with important people in their lives, including family, peers, teachers, and so on. At the next level is the mesosystem, which forms when people from different microsystems containing the developing child interact. For the current study, we situate our analysis of TSRs and involvement in peer aggression at the level of the mesosystem, as classrooms provide numerous opportunities for mesosystems comprising teachers and peers to form and to mutually influence one another.

From this bioecological perspective, social referencing theory is a useful explanation of one potential mechanism by which teachers may influence students’ behaviours in their peer group, including peer aggression involvement (Hughes & Chen, 2011). In brief, social reference theory suggests that students use their teachers as a social reference to inform how students should treat their peers, effectively mirroring to a peer the qualities they perceive in that peer's relationship with the teacher. While some longitudinal research suggests bidirectional links between TSR quality and involvement in peer aggression (e.g., Holfeld & Leadbeater, 2017; Serdiouk et al., 2016), it seems that the teacher is the driving influence in the association between the two constructs. (e.g., Hendrickx et al., 2016). This teacher influence seems to be especially true for conflictual interactions between the teacher and student where students are more likely to dislike their peer if they perceive that their teacher dislikes them.

Attachment Theory (Bowlby, 1969) has also been frequently used to describe how TSRs affect levels of peer aggression involvement. Attachment theory describes the psychosocial development of children within the context of caregiving relationships and offers a useful framework to understanding the quality of relationships that can form between students and teachers. According to Attachment Theory, the dynamics of early relations with caregivers—specifically the levels of sensitivity and attunement they direct toward the child—lead the child to develop an internal working model of relationships (including expectations and core beliefs) that roughly divide between models of secure and insecure attachment. This implies, and research offers some supportive evidence, that the quality of TSRs is related, at least in part, to parental attachment history (Sabol & Pianta, 2012). However, the theory also suggests that non-parental relationships, including those that form between teachers and students, can themselves form into attachment relationships that are at least partially independent from parental attachment histories, and thereby contribute to shaping or revising children's internal relational models (Davis, 2003). Revised models that favour secure attachment in the TSR can then serve to guide behaviour that subsequently contributes to the formation of high-quality peer relationships (Gest & Rodkin, 2011).

Moderating variables

Research has documented multifaceted trends in the relationship between peer aggression involvement and TSR closeness as a function of students’ developmental age. For example, a review by Berger (2007) outlines several studies that found physical bullying decreases in high school, but more covert forms of bullying (e.g., verbal, social, and cyberbullying) increase with age, particularly during middle school years (Grade 6–9) (Due et al., 2005). Regarding TSR closeness, literature indicates that younger students tend to view their relationship with their teachers as more positive, compared to older students (Havik, 2017; Serdiouk et al., 2016). It is also well established that as children age their relationships with peers take increasing importance in their lives (Brown & Larson, 2009). Additionally, students in upper grades (e.g., Grade 7 and above) often have several teachers with whom they interact throughout the day rather than a single homeroom teacher. This too, may influence the observed association between involvement in peer aggression and TSR quality. It is for these reasons that school level, a proxy for developmental age, was included as a moderator in our analyses.

Multiple informants are often used in measuring peer aggression and TSR quality (Richard et al., 2011; Runions & Shaw, 2013; Troop-Gordon & Kopp, 2011) given the reporting limitations associated with each type of informant. For example, social desirability may influence self-reports of peer aggression involvement (Volk et al., 2017). It is also known that some systematic reporting differences exist between informants when reporting involvement in peer aggression (Demaray et al., 2013; Stockdale et al., 2002), which is often explained in terms of varying opportunities that different informants have to witness actual peer aggression (Rupp et al., 2018). Given these trends, we decided to include informant type as a moderator in our analysis.

As the final moderator, we will assess the influence of type of measure. Many of the measures used to investigate peer aggression involvement and TSR quality are well-established and extensively validated (Kimberlin & Winterstein, 2008). However, many studies nonetheless included newly created measures that have very limited validity and reliability data compared to the established measures. Consequently, we decided to compare effect sizes based on well-established measures versus novel measures of our key variables.

The current study

The objective of our meta-analyses is to quantitatively synthesize the size and direction of the association between TSR closeness and students’ involvement in peer aggression, namely as perpetrators and as victims. We expected that students who perpetrate against their peers or who are victimized will also have relationships with their teachers that are low in closeness. Peer perpetration and victimization were investigated independently. We divided the analyses in this way since the peer aggression literature routinely and reliably divides children into various roles related to peer aggression involvement (Yang & Salmivalli, 2013). Scholarship also documents unique outcomes related to peer aggression involvement for students who are victimized and students who perpetrate against others (Ttofi & Farrington, 2008). Additionally, we examined the impact of moderating variables (e.g., school level, informant type, and measure type) on the association between peer aggression involvement and TSR closeness. We expected that all three moderator analyses would help explain the heterogeneity across studies. Regarding the school level moderator, we anticipated a stronger association between TSR closeness and peer aggression involvement for younger students compared to older students given that younger students are more dependent on adults in their lives and spend greater amounts of the school day with a single teacher (Wang et al., 2015). Given the limited research indicating how informant or measure type may influence the correlation between involvement in peer aggression and TSR closeness, these two analyses are exploratory in nature.

Study selection and screening

A systematic literature search was performed in September 2019 to identify all relevant literature. Four online databases were accessed: ERIC, PsycINFO, Education Source, and ProQuest (for theses and dissertations only). A combination of Medical Subject Heading (MeSH) terms and key words were used to reliably access relevant literature. Some examples of MeSH terms include: relational aggression and teacher student relationship. Some examples of key words used include: peer harass* and teacher adj2 student adj2 relation* (see supplemental material for a full description of the systematic literature search strategy). Backward and forward referencing searches were also completed to ensure all possible studies on the topic were identified. At the end of the search process, there were a total of 816 unique research reports identified for this study. These reports were then uploaded onto an online systematic review manager (i.e., Covidence) for screening.

To be included in the meta-analyses, all studies must have included a quantitative and continuous measure of student peer aggression involvement, either perpetration or victimization, from a student (self), teacher, and/or peer informant that was completed in the school context. Peer aggression involvement, in the context of this study, included aspects of bullying and victimization, as well as other forms of peer aggression (e.g., reactive aggression) that do not necessarily fit within the confines of the traditional bullying definition. We included all forms of peer aggression involvement, including physical, verbal, vandalism, and cyber, that occurred in the context of an interpersonal interaction, and we excluded measures of aggression that targeted objects or animals. An important role identified in the peer aggression literature is the bully-victim role, which comprises students who both aggress against others and are victimized themselves. Due to key differences between students who both perpetrate and are victimized and students who only perpetrate against others or are purely victimized, it would be necessary to examine these roles in separate analyses, however, insufficient number of studies that reported on this specific group of students prevented us from doing so. Additionally, all studies must also have included a quantitative and continuous measure of TSR closeness from either the student (self), teacher, and/or peer informant that was completed in the school context. Measures that assessed positive aspects of TSRs were included (e.g., Aceves et al., 2009; Espelage et al., 2014), such as aspects of support, trust, warmth, and mutual liking. Finally, all studies needed to include individual-level statistics, have study participants who were in grades k-12, and include a correlation coefficient effect size or other statistic that could be transformed into one.

The first author was the primary reviewer during the screening process. When it was not clear whether a study fit within the predetermined inclusion/exclusion criteria, a consensus decision was made following a discussion between the two authors. After all studies were screened and assessed according to the inclusion and exclusion criteria, a total of 66 studies remained. See Figure 1 for a description of the selection and screening process.

OPEN IN VIEWER

Figure 1.

PRISMA diagram of study selection process.

Study coding process

See Table 1 for a list of all included studies and their characteristics. The first author coded the entire set of 66 studies. A second coder, a graduate student completing a master's degree in education and who had no knowledge of the study purpose, was trained on how to identify and code each study characteristic. The second coder coded 19 studies (>28%) of the final set of studies to ensure inter-rater reliability. Inter-rater agreement was computed by matching the coding of both raters and calculating the number of agreements against the total number of comparisons. Overall inter-rater agreement was excellent at 97% percent.

OPEN IN VIEWER

Table 1.

Characteristics of studies included in the meta-analyses.

		Sample Characteristics					Measurement Characteristics			Effect Size
Authors	N	Country	School Level	Gender	Ethnicity	PA Measure	PA Informant	TSRclos Measure	TSR Informant	r
Aceves et al., 2009	148	USA	Middle; Secondary	49.3% female	76 (51.5%) Latino, 36 (24.2%) Asian American, 28 (18.9) African American, 8 (5.3%) other (including Caucasian, Middle Eastern, or mixed ethnicity)	Reactive-Proactive Aggression Questionnaire (RPQ; Raine et al., 2006) to measure pro-active peer aggression	Self	Trust In and Respect For Teacher Scale (Battistich et al., 2004) to assess students appraisals of TSRQ	Self	-.277 (P)
Akiba et al., 2010	2970	Japan	Middle; Secondary			Created own measure to assess victimization and bullying behaviour	Self	Created own measures to assess Teacher Bonding, Student guidance, and Instructional support	Self	-.113 (P).107 (V)
Bacchini et al., 2009	734	Italy	Middle; Secondary	50.1% female		Olweus Bully victim questionnaire (Olweus 1986 1996) to assess frequency of victimization and perpetration	Self	Created own measure to assess student perceived relationship with teacher, and if teacher has favourites	Self	-.090 (P).070 (V)
Bae et al., 2018	313	South Korea	Middle; Secondary	51.6% female	148 (59%) of sample were of ethnic minority (at least 1 foreign born parent); Other half were of Korean ethnicity	Olweus Bully/Victim Questionnaire (modified version; Harel-Fisch et al., 2011; Olweus, 2007) to assess frequency of victimization and perpetration	Self	Student's Perceived Relationship with Teachers to assess teacher support and care of student (Harel-Fisch et al., 2011)	Self	-.273 (P)-.355(V)
Bardack & Obradovic, 2017	89	USA	Elementary	52% female	59% minority status (25% Hispanic/Latino, 19% Asian, 12% Multiracial/ other, 3% Black)	MacArthur Health and Behaviour Questionnaire (Armstrong & Goldstein, 2003) to assess relational aggression	Teacher	Teacher-child closeness scale (Armstrong & Goldstein, 2003) adopted from STRS to assess teacher's perceptions of closeness with individual child	Teacher	-.020 (P)
Beran, 2009	794	Canada	Middle; Secondary	49.8% female		Survey developed for National Longitudinal Survey of Children and Youth (Statistics Canada, 1999) to measure frequency of victimization and bullying behaviour	Self	Survey developed for National Longitudinal Survey of Children and Youth (Statistics Canada, 1999) perceived support from teachers and fair treatment	Self	-.288 (P)
Boulton et al., 2012	364	United Kingdom	Elementary	46.7% female		Peer nominations with items derived from previous studies (e.g., Boulton & Smith, 1994) to assess variety of bullying experiences	Peer	Created own scale to measure TSR on positive aspects only	Self	-.141(V)
Boyesen & Bru, 1999	1071	Norway	Middle; Secondary			Created own measure to assess harassment of others and exposure to harassment by peers	Self	Created own measure to assess student perceptions of teacher support	Self	-.138 (P)-.048(V)
Brendgen et al., 2011	434	Canada	Elementary	50.2% female	87% European decent, 3% African descent, 3% Asian descent, 1% Native North Americans. 6% did not disclose.	Peer nominations with items from Victimization subscale of modified Peer Nomination Inventory (Perry et al., 1988) and items inspired by Preschool Behaviour Questionnaire (PBQ; Behar & Stringfield, 1974) to assess victimization and bullying experiences of classmates	Peer	Teacher-Child Relationships Scale (STRS; Pianta et al., 1995), to assess closeness and conflict in TSR	Teacher	-.255 (P)-.100(V)
Bru et al., 2002	3834	Norway	Middle; Secondary			Created own measure to assess harassment of others	Self	Scale developed by Center for Behavioural Research to assess student perception teacher support	Self	-.085 (P)
Cho & Wooldredge, 2017 (South Korea)	2671	South Korea	Middle; Secondary		Mostly homogenous racial make-up of country's youth population	No specific scale mentioned to assess frequency of bullying	Self	No specific scale mentioned to assess attachments to teachers	Self	-.037 (V)
Cho & Wooldredge, 2017 (USA)	4990	USA	Middle; Secondary		22% Non-white	No specific scale mentioned to assess frequency of bullying	Self	No specific scale mentioned to assess attachments to teachers	Self	-.404 (V)
Cornell et al., 2015	39 364	USA	Middle; Secondary	51.7% female	52.4% White, 18.2% Black, 12.8% Hispanic, 3.4% Asian, 1.6% American Indian or Alaska Native, .5% Native Hawaiian or Pacific Islander, 15.6% identifying self as more than one race	No specific scale name mentioned to assess frequency of victimization; General Victimization scale (Gottfredson, 1999) to assess general victim experiences	Self	Derived from Learning Environment Scale (Austin & Duerr, 2005) and Willingness to Seek Help scale (Bandyopadhyay et al., 2009) to assess perceived supportiveness of teacher	Self	-.354 (V)
Cortes & Kochenderfer-Ladd, 2014	278	USA	Elementary	51.4% female	46.3% Hispanic or Mexican American, 43.5% White, 10.2% Black and Other	Created own peer rating items to assess classmates picking on others;Multi-source Peer Victimization Inventory to measure victimization experiences	Peer; Self	Student-Teacher Relationship Scale (STRS; Pianta, 2001) to assess closeness of TSR	Teacher	-.192 (P)-.035 (V)
Datta et al., 2017	56 508	USA	Middle; Secondary	48.3% female	15.8% Hispanic or Latino, 49.4% White, 18.1% Black, 6.8% Asian, 2.2% American Indian or Alaskan Native, 1.0% Native Hawaiian or Pacific Islander, 22.5% Multiracial	No specific scale mentioned to assess frequency of being bullied	Self	Derived from the Authoritative School Climate Survey (Cornell et al., 2013) to assess supportiveness of TSR	Self	-.151 (V)
Davidson & Demaray, 2007	355	USA	Middle; Secondary	53% female	97% Caucasian (344), 2% Hispanic (7), 0.8% African American (3), .3% Asian American (1)	Bully-Victimization Scale (BVS; Reynolds, 2003) to assess victimization and bullying experiences	Self	Child and Adolescent Support Scale (CASSS; Malecki et al., 2000) to measure social support from teachers	Self	-.084 (P)-.097 (V)
Di Stasio et al., 2016	678	Canada	Middle	51% female	55.2% European descent, 4.4% South Asian, 1.4% Southeast Asian, 1.7% African American, 1% Aboriginal, .7% West Asian, .7% Latin American, 23.3% other, 11% no response	Illinois Bully Scale (IBS; Espelage & Holt, 2001) to assess frequency of bullying, victimization, and fighting	Self	Student's Classroom Environment Measurement (SCEM; Midgley et al., 1991), subscale of Teacher/student relations to assess fair treatment and care for students	Self	-.140 (P)-.260 (V)
Doty et al., 2017	121 311	USA	Middle; Secondary	49.8% female	74% White, 6.9% Multiracial, 7.3% Hispanic, 5.3% Black, 5.6% Asian or Pacific Islander, 1.2% American Indian/Alaskan Native	California Healthy Kids survey (California Department of Education, 2010) to assess frequency of bullying perpetration and victimization	Self	Created for the Minnesota Student Survey to assess perceived connectedness with teacher	Self	-.224 (V)
Doumas & Midgett, 2019	110	USA	Elementary	59.1% female	62.7% Caucasian, 9.1% Hispanic, 9.1% African American, 6.4% Asian, 0.9% Pacific Isalnder, 11.8% other	Olweus Bullying Questionnaire (Olweus, 1996) to assess bullying perpetration and victimization	Self	Psychological Sense of School Membership (PSSM; Goodenow, 1993), used subscale of caring relationships to assess quality of TSR	Self	-.406 (V)
Elledge et al., 2016	361	USA	Elementary	53.9% female	61.1% Caucasian, 25.4% Hispanic, 13.5% other racial/ethnic groups	Adapted School Experiences Questionnaire (SEQ; Kochenderfer-Ladd, 2004) to assess peer victimization; No specific scale mentioned to assess teacher's ratings of victimization frequency;Modified version of Revised Class Play (Masten et al., 1985) to assess bullying and victimization of classmates	Self; Teacher; Peer	Peer nomination procedure (Hughes et al., 2001) to assess peer perceptions of classmate TSR quality	Peer	-.216 (V)
Elsaesser et al., 2013	5106	USA	Middle; Secondary	49.2% female	52.1% African American, 21.2% Hispanic, 17.3% Caucasian, remaining as American Indian or Asian	Problem Behaviour Frequency Scales (Farrell et al., 2000) to assess relational aggression and victimization	Self	Vessels’ School Climate Survey (Vessels, 1998) to assess fair treatment and criticisms from teachers	Self	.100 (P)-.255 (V)
Ertesvag, 2016	1571	Norway	Middle; Secondary	50.8%, 1.4% did not report gender		No specific scale (previously validated by Roland, 1999) mentioned to assess various types of bullying	Self	No specific scale mentioned to assess teacher emotional support and teacher instructional support scale	Self	-.110 (P)
Espelage et al., 2014	3616	USA	Middle; Secondary	48% female	Illinois: 33.5% Black, 33.1% Latino, 1.0 Asian, 20.2% White, 11.7% Biracial.Kansas: 35.2% Latino, 30.4% White, 17.5% Black, 4.2% Asian, 12.8% Biracial.	Illinois Bully Scale to measure bullying perpetration;Illinois Fighting scale to measure frequency of physical aggression;Illinois Victimization Scale to assess frequency of peer victimization (all Espelage & Holt, 2001)	Self	Subscale of School Environment Survey, adapted from Colorado Trust's Bullying Prevention Initiative (Csuti, 2008), to assess positive teacher-staff-student interactions	Teacher	-.611 (P)-.497 (V)
Flaspohler et al., 2009	4331	USA	Elementary	49.2% female		Bully/Victim Questionnaire (Olweus, 1996) to assess frequency of peer victimization	Self	Teacher subscale of Child and Adolescent Social Support Scale (CASSS; Malecki et al., 2000) to assess social support from teachers	Self	-.147 (V)
Gyllborg, 2013	116	USA	Elementary	54.3% female	90.4% (103) White, 5.3% (6) Hispanic, .9% Black (1), 3.5% Asian (4)	Relational aggression subscale of Children's Social Behaviour Scale (CBS-S; Crick & Grotpeter, 1995) to measure relational aggression; Relational victimization subscale from Children's Social Experiences Questionnaire to assess relational victimization experiences	Self	Short form of Student-Teacher Relationship Scale (Pianta et al., 1995) to assess conflict and closeness in TSR	Teacher; Self	-.143 (P).097 (V)
Havik, 2017	1571	Norway	Middle; Secondary	50.8% female		Scale developed by Roland and Idsøe (2001) to assess various types of peer victimization	Self	No specific scale mentioned to assess teacher emotional support and teacher instructional support	Self	-.188 (V)
Herrero et al., 2006	973	Spain	Middle; Secondary	52.8% female		Scale taken from Mynard and Joseph (2000) to measure various forms of peer victimization	Self	Created own measure to assess quality TSR and degree of non-conflict	Teacher	-.175 (P)-.060 (V)
Idsoe et al., 2008	2083	Norway	Middle; Secondary	51.5% female		Developed at Centre for Behavioural Research (Roland & Idsøe, 2001) to assess frequency of bullying behaviour	Self	Adopted from School Environment Questionnaire developed by Centre for Behavioural Research to assess teacher emotional support and teacher autonomy	Self	-.326 (P)
Iimori, 2002	817	USA	Middle; Secondary	48.7% female	90.3% Caucasian, 1.8% Asian, 1.1% African-American, 1.1% Hispanic, 1.0% Native American, 4.7% other.	No specific scale indicated to assess peer aggression; No specific scale mentioned to assess peer victimization experiences	Self	No specific scale mentioned to assess teacher support for student	Self	-.199 (P)-.147 (V)
Jia et al., 2018	747	China	Middle; Secondary	48.9% female		Adolescent Peer Victimization Questionnaire (Li et al., 2017) to assess frequency of peer victimization	Self	Adapted from Inventory of Parent and Peer Attachment (Jia et al., 2017) to assess closeness in TSR	Self	-.224 (V)
Khoury-Kassabri, 2011	4495	Israel	Elementary	54% female	Two ethnic groups: Arab schools and Jewish schools. Higher response rate from Arab schools (58.8%).	Adapted from California School Climate Survey (Furlong et al., 2005; Rosenblatt & Furlong, 1997) to measure frequency of peer victimization	Self	The School Climate Scale (Benbenishty, 2003) to assess teacher support	Self	-.105 (V)
Kremer, 2010	153	USA	Elementary	47.1% female	42% Hispanic, 46% White, 2% African American, 4% Native American, 1% Asian, 5% Biracial	This Child's Interactions with Peers Scale (Ladd & Kochenderfer-Ladd, 2002) to assess frequency of peer victimization	Teacher	Truncated version of Student-Teacher Relationship Scale (Pianta, 1994), to assess closeness, and conflict	Teacher	-.151 (V)
Kuhns, 2008	697	USA	Elementary	47.4% female	34.1% White, 37.4% Hispanic, 23.2% African-American, 3.6% Asian/Pacific Islander, .3% Native American/Alaskan Native, 1.5% Other	Peer nomination procedure to assess aggressive behaviour of classmates	Peer	Teacher Student Relationship Inventory (TSRI; Hughes et al., 2001) to assess teacher support and conflict;Peer nomination procedure to assess teacher support of classmates	Teacher; Peer	-.266 (P)
Lam et al., 2017	454	Hong Kong	Middle; Secondary	49.4% female	All Chinese	Several scales to assess frequency of peer aggression and victimization (all scales Hill & Werner, 2006)	Self	Caring Adult Relationships in School Scale in California Healthy Kids Survey (WestEd, 2000) to assess teacher support for relatedness	Self	-.277 (P)-.218 (V)
Leadbeater et al., 2014	800	Canada	Elementary	51% female	95% European Caucasian, 2% Aboriginal, 1% Asian, 1% mixed ethnicity. Less than 1% Hispanic, Indo-Canadian, or African Canadian.	Social Experience Questionnaire (SEQ, Crick & Grotpeter, 1996) to assess frequency of peer victimization experiences	Self	Student-teacher relations aspect of School Climate Survey (Haynes et al., 1993) to assess closeness and trust in TSR	Self	-.277 (P)
Lohbeck & Petermann, 2018	332	Germany	Middle; Secondary	54.5% female		Created new scales to assess various forms of cyber-victimization	Self	Adapted from Teacher-Student Relationship Questionnaire of PISA (Kunter et al., 2002) to assess quality and caring in TSR	Self	-.125 (P)
Longobardi et al., 2018	435	Italy	Middle; Secondary	48.8% female	4.4% of sample consisted of 1st of 2nd generation immigrants.	Olweus Bully/Victim Questionnaire (OBVQ; Olweus, 1996) and Participant Role Scale (PRS; Jungert et al., 2016) to assess involvement in bullying behaviour and victimization	Self	Student Perception of Affective Relationship with Teacher Scale (SPARTS; Koomen & Jellesma, 2015) to assess closeness, conflict, and negative expectations within TSR	Self	-.139 (P)
Lucas-Molina et al., 2015	1864	Spain	Elementary	50.7% female		Adapted from another study (Defensor del Menor de la Comunidad de Madrid, 2006) to assess frequency of victimization and peer aggression	Self	Adapted from Benbenishty and Astor (2005) to assess student perceived teacher support in TSR;Student-to-teacher aggression scale (Mendoza, 2006), to assess frequency of aggressive behaviours towards teacher by student;Direct and indirect teacher-to-student aggression scale (Mendoza, 2006) to assessed frequency of teacher aggressive behaviour towards student	Self	-.020 (P)-.095 (V)
Mann et al., 2015	7084	Iceland	Middle; Secondary	51.5% female		Developed by Icelandic Institute for Educational Research (IER) and ICSRA to assess frequency of bullying perpetration and victimization	Self	No specific scale indicated to assess quality of TSR	Self	.075 (P).016 (V)
Marachi et al., 2007	8404	Israel	Middle; Secondary	50% female		Adapted from California School Climate Survey (Furlong, 1999) to assess presence of victimization	Self	Adapted from California School Climate Survey (Furlong, 1999) to assess teacher support	Self	-.430 (V)
McAuliffe et al., 2009	127	USA	Elementary	49.6% female	32.3% European American, 31.5% African American, 25.2% Hispanic/Latino, 3.2% another race. 6.3% more than one race, 1.6% no response	Peer nomination procedure (Crick & Grotpeter, 1995) to assess peer aggression in classmates	Peer	No specific scale indicated to assess teacher liking for each individual student	Teacher	-.604 (P)
Meehan et al., 2003	140	USA	Elementary	33.6% female	37% Caucasian, 41% African American, 22% Hispanic.	Modified version of Revised Class Play Method (Masten et al., 1985) to assess relational aggressive behaviour in classmates	Peer	Network of Relationships Inventory (NRI; Furman & Buhrmester, 1985) to assess aspects of support from teacher including	Self	.030 (P)
Moore et al., 2017	761	USA	Middle/Secondary	25.6% female	11.4% (133) Other, 13.1% (153) Black, 14.1% (165)White, 10.9% (128) Mixed, 50.5% (590) Hispanic	California Healthy Kids Survey (CHKS; WestEd, 2000) to assess frequency of victimization	Self	California Healthy Kids Survey (CHKS; WestEd, 2000) to assess caring relationships with teacher	Self	.121 (V)
Murray-Harvey & Slee, 2010	888	Australia	Middle; Secondary	49% female	No specific data on sample - population typically Anglo-European, with minority of Asian, Middle-Eastern, and Indigenous Australian students.	Your Life at School questionnaire to assess frequency of bullying perpetration and victimization	Self	Your Life at School questionnaire to measure presence of stressfulness and supportiveness in TSR	Self	-.321 (P)-.215 (V)
Pabian & Vandebosch, 2016	2128	Belgium	Middle; Secondary	50.5% female	94.8% Belgian nationality.	No specific scale mentioned to assess frequency of cyber-bullying, cybervictimization, traditional bullying, and traditional bullying victimization	Self	Adapted Dutch language school commitment scale (Murdock & Phelps, 1973) to assess student bonding to teacher	Self	-.094 (P).031 (V)
Raskauskas et al., 2010	1168	New Zealand	Elementary	52% female	Although no data collected here, sample represented ethnic makeup of New Zealand school pop.:70% European, 15% Maori, 10% Asian, 5% Pacific nations.	Peer Relations Questionnaire (PRQ; Rigby 1997) to assess prevalence of bullying and victimization	Self	Student-teacher relationships subscale from Safe Communities -Safe Schools survey (Elliott et al., 2000) to assess quality of TSR	Self	-.203 (P)-.030 (V)
Reavis et al., 2010	218	USA	Elementary	56% female	64% Caucasian	Peer nomination procedure (Coie et al., 1982) to assess peer victimization	Peer	Student-Teacher Relationship Scale (STRS; Pianta, 2001) to assess conflict and closeness in TSR	Teacher	-.030 (V)
Richard et al., 2011	18 222	France	Middle; Secondary	52% female	85% Caucasian	Scale developed by Dauphin and Trosseille (2004) to assess peer victimization	Self	Scale developed by Dauphin and Trosseille (2004) to measure quality of TSR	Self	-.202 (P)
Rigby & Bagshaw, 2003	3551	Australia	Middle; Secondary			No specific scale indicated (e.g., Rigby & Bagshaw, 2001) to assess frequency of indirect bullying perpetration and victimization	Self	Created own scale for study to assess perceived helpfulness and fair treatment of teachers	Self	-.141 (P)-.070 (V)
Roth et al., 2010	725	Israel	Middle; Secondary	50% female		Scale developed by Rolider et al. (2001) to assess frequency of peer aggression	Self	Created own measure from Assor et al. (2002) Roth et al. (2007) and the Learning Climate Questionnaire (Williams & Deci, 1996) to assess student perception of teacher supporting student autonomy	Self	-.266 (P)
Rucinski, 2015	526	USA	Elementary	53.8% female	44.5% (234) Hispanic, 18.1% (95) Black, 11.8% (62) White, 25.5% (134) Other racial-ethnic groups (approx. 95% of Other group identified as Asian/Pacific Islander).	Aggression Scale (Orpinas and Frankowski, 2001) to assess frequency of peer aggression	Self	Student-Teacher Relationship Scale (STRS; Pianta, 2001) to assess closeness and conflict of TSR;Drawn from Classroom Assessment Scoring System to assess trust and emotional support (Pianta et al., 2010)	Teacher; Self	-.131 (P)
Rudasill et al., 2017	1364	USA	Elementary	48.3% female, 1 missing	80% (1097)White, 13% (176) Black, 7% (91) Hispanic, Asian American, or American Indian.	Adapted from Child Behaviour Scale (Ladd & Profilet, 1996), Peer Victimization Scale (Kochenderfer & Ladd, 1996), and Relational Aggression scale (Crick & Grotpeter, 1996) to assess bullying perpetration and victimization	Teacher	Student-Teacher Relationship Scale (STRS; Pianta, 2001) to assess closeness and conflict of TSR	Teacher	-.100 (P)-.090 (V)
Runions & Shaw, 2013	377	Australia	Elementary	50.1% female	Most spoke English in home. Non-English speakers were representative of immigration to Western Australia.	Derived from the Social Behaviour Questionnaire (SBQ; Tremblay et al., 1991) to assess physical peer aggression	Teacher	Student-Teacher Relationship Scale (STRS; Pianta, 2001) to assess closeness and conflict in TSR	Teacher	-.192 (P)
Serdiouk et al., 2016	1568	USA	Elementary	48.8% female	47% European American, 35% African American, 10% Hispanic, 4% Other	Olweus Victim Questionnaire (Olweus, 1996) to assess frequency of peer victimization	Self	Created own scale resembling closeness subscale of the Student-Teacher Relationship Scale (STRS; Pianta & Steinberg, 1992) to assess closeness and avoidance in TSR	Teacher	-.070 (V)
Sulkowski & Simmons, 2017	539	USA	Middle; Secondary	51% female	43% (236) Hispanic/Latino, 27% (145) White/Caucasian, 8% (41) Black/African American, 4% (19) Asian American, 2% (8) Native American/American Indian, 1% (4) Pacific Islander, 4% (19) Other.	Peer Victimization scale on Revised Peer Experiences Questionnaire (RPEQ; Prinstein et al., 2001) to assess frequency victimization	Self	Teacher-Student Relationships subscale on School Climate Measure (SCM; Zullig et al., 2010) to assess quality of TSR	Self	-.424 (P)
Tanigawa, 2009	544	USA	Middle; Secondary	56% female	40% Hispanic/Latino, 29% White, 20% Multi-ethnic, 5% Asian, 2% Black/African American, <1% American Indian or Alaskan Native, <1% Other.	Victim scale of the California Bully/Victim Scale (Furlong et al., 2007) to assess frequency of peer victimization	Self	Child and Adolescent Social Support Scale (CASSS; Malecki et al., 2000) to assess perceived social support from teachers	Self	-.472 (V)
Thornberg et al., 2017	899	Sweden	Elementary	48.4% female	Majority Swedish background, 16% had immigrant background.	Created own measure (Thornberg et al., 2017) to assess prevalence of peer victimization	Self	Adopted the Class relational climate scale (Thornberg et al., 2017) to assess quality of TSR	Self	-.388 (V)
Troop-Gordon & Kopp, 2011	410	USA	Elementary	52.9% female	87.1% European American, 1% African-American, 1.2% Asian-American, 4.6% Native American, 1% Hispanic, 5.1% mixed ethnic background.	Peer rating derived from Ladd and Kochenderfer-Ladd's (2002) scales to assess frequency of peer aggression and victimization	Peer	Student-Teacher Relationship Scale (STRS; Pianta & Steinberg, 1992) to assess closeness and conflict	Teacher	-.123 (P)
Troop-Gordon & Kuntz, 2013	352	USA	Elementary	52.8% female	88% Caucasian, .9% Asian-American, .6% African-American, 1.4% Hispanic, 4.6% Native American, 4.6% mixed or other ethnicity.	Adapted from Multi-Informant Peer Victimization Inventory (Ladd & Kochenderfer-Ladd, 2002) to assess peer harassment	Self	Items taken from the Student-Teacher Relationship scale (STRS; Pianta & Steinberg, 1992) to assess closeness and conflict in TSR	Self	-.020 (V)
Unal & Cukur, 2011	3742	Turkey	Middle; Secondary	48% female		No specific measure indicated to assess frequency of victimization	Self	Adopted from Richmond Youth Project (Hirschi, 1969) to assess student attachment to teacher	Self	-.192 (V)
Valle et al., 2018	426	Brazil	Middle; Secondary	58% female		School Violence Scale (Stelko-Pereira et al., 2010) to assess prevalence of bullying perpetration and victimization	Self	Teacher-Student Relationship Scale (Lamas et al., 2013) to assess closeness and conflict in TSR	Self	-.310 (P)-.266 (V)
Wang et al., 2015	435	USA	Middle; Secondary	56.8% female	71.3% Caucasian, 6% African American, 4.1% Asian American, 3.4% Hispanic, 1.4% Eastern European.	Verbal and Physical Bullying Scale (VPBS; Swearer et al., 2008; Radliff et al., 2016) to assess bullying perpetration	Self	School Climate Survey Revised Edition - Elementary and Middle School Version (SCS-ESV; Emmons et al., 2002) to assess teacher-student relationship quality	Self	-.208 (P)
Wang et al., 2018	1148	China	Elementary	45% female		Delaware Bullying Victimization Scale (Bear et al., 2011) to assess perceived bullying victimization	Self	Delaware School Climate Survey (Bear et al., 2011) to assess teacher-student relationship quality	Self	-.239 (V)
Wang et al., 2016	1075	USA	Elementary	50% female	48.1% White/European American, 28.7% Black/African American, 11% Hispanic, 7.3% Asian, .4% American Indian, 4.5% multiracial.	No specific scale indicated to assess prevalence of peer victimization	Self	No specific scale indicated, assessed teacher preference for each child, 1 item (5-point scale);No specific scale indicated, assessed student perceived relationship quality with teacher,	Teacher; Self	-.100 (V)
Yang, 2015	25 896	USA	Middle; Secondary	51.5% female	45.4% Caucasian, 24.6% African American, 14.6% Hispanic/Latino, 3.55% Asian, 11.9% Other	Delaware Bullying Victimization Scale - Student (DBCS-S; Bear et al., 2014) to assess frequency of bullying victimization	Self	Teacher-Student Relationships subscale of Delaware School Climate Survey - Student (DSCS-S; Bear et al., 2014), assessed quality of interactions between teachers and students, 4 items (4-point scale)	Self	-.141 (V)
Yeung & Leadbeater, 2010	664	Canada	Middle; Secondary	51.5% female	86% Caucasian, 4% Asian, 2% Aboriginal, 1% Hispanic, 1% African, 1% East Indian, 1% Middle Eastern, 4% Other	Relational victimization and physical victimization subscales of Social Experiences Questionnaire (SEQ; Crick & Grotpeter, 1995) to assess frequency of peer victimization	Self	No specific scale indicated, assessed teacher emotional support and reciprocity as perceived by student, 3 items (5-point scale)	Self	-.151 (V)
Zanden et al., 2015	781	Netherlands	Elementary	49.9% female		Bullying others' and 'victim of bullying' subscale of Olweus Bully/Victim Questionnaire (Olweus, 1989) to assess frequency of bullying perpetration and victimization	Self; Self	Questionnaire on Teacher Interaction (QTI; Wubbels et al., 2006) to assess aspects of closeness and conflict	Self	-.133 (P)-.157 (V)

Note. PA = Peer aggression; TSRclos = Teacher-student relationship closeness; P = Peer perpetration; V = Peer victimization.

Aggregation of effect sizes

We aggregated across multiple measures of peer aggression involvement and TSR quality that were reported within a single study. This was required for a total of 35 studies. Aggregation was done by summing the various effects from each measure and dividing it by the number of effect sizes/measures, providing an average effect size across measures used. For the longitudinal studies (total of 16 studies), effect sizes were aggregated if they corresponded with timepoints occurring within the same academic year. When the timepoints of longitudinal studies fell in different academic years (e.g., timepoint 1  =  Grade 1, timepoint 2  =  Grade 2), only the first timepoint was extracted. This was done to minimize the impact of development on involvement in peer aggression and TSR quality as students age (Havik, 2017). Aggregation for these effect sizes was done by summing the total number of effect sizes occurring in the same academic year and dividing it by the number of effect sizes/time points. One intervention study was included in the final set (Elsaesser et al., 2013), and only the cross-sectional data collected prior to the intervention was used to eliminate effect size bias. A small subset of the unique studies reported on the same sample of students. In these cases, when the data were unique (e.g., one study reported on perpetration while the other reported on victimization; see Troop-Gordon & Kopp, 2011, and Troop-Gordon & Kuntz, 2013), both studies with their respective data were included in the final set. If the data in the two studies were exactly the same, only one study's data were included.

Study characteristics

With our set of 66 research reports we calculated a total of 90 independent effect sizes. Of the 66 included studies, 59 (89.4%) were published studies, and seven (10.6%) were unpublished studies (two master's theses and five doctoral dissertations). Publication dates of the set of studies ranged from 1999–2019. A total of 352, 376 student participants are represented in the present meta-analyses. The mean sample size is 5244 with sample sizes ranging from 89 to 121,311 participants (the largest, a US sample; Doty et al., 2017).

The studies in our dataset represent 20 countries from five continents (see Table 2 for a breakdown of participants across geographical regions). The majority of studies (>80%) reported roughly equal proportions of males and females in their study. Approximately 40% of studies included ethnically heterogeneous samples, and just over 20% of the studies reported ethnically homogeneous samples (i.e., indicated by over 80% of one ethnic background). Roughly 37% of the studies did not provide any ethnicity information about study participants. Students of all ages were included in the present meta-analyses from kindergarten to Grade 12, with 40% of studies including elementary school-aged children, and the remaining 60% including middle/intermediate and secondary school-aged children.

OPEN IN VIEWER

Table 2.

Breakdown of included studies and participants by geographical region.

Geographical Region	Number of Studies	Number of Participants
By Country
Australia	3	4816
Belgium	1	2128
Brazil	1	426
Canada	5	3370
China	3	2349
France	1	18 222
Germany	1	332
Iceland	1	7084
Israel	3	13 524
Italy	2	1169
Japan	1	2970
Netherlands	1	781
New Zealand	1	1168
Norway	5	8559
South Korea	2	2984
Spain	2	2837
Sweden	1	899
Turkey	1	3742
United Kingdom	1	364
USA	31	271 953
By Continent
North America	36	275 323
South America	1	42 6
Europe	17	46117
Asia	9	21 827
Australia	4	5984

Three quality assessment criteria, extracted from the Quality Assessment Tool for Studies with Diverse Designs (QATSDD; Sirriyeh et al., 2011) were coded to determine the amount and specificity of information for each indicator of reporting quality. After careful deliberation of other quality assessment tools, the QATSDD was used since it contained items that were highly relevant to the diverse but primarily observational and correlational nature of the included studies. No other quality assessment tools (e.g., Cochrane risk of bias tool, Newcastle-Ottawa Scale) were better suited for our specific set of studies. The three quality assessment items that were coded included clear description of research setting (e.g., target population and clear research questions), clear description of data collection procedure (e.g., when and where data was collected), and statistical assessment of reliability and validity for measurement tools. These three items were selected since they queried critical and over-arching reporting standards observed in high quality studies. Fifty-one studies (77.3%) satisfied all three quality assurance criteria, 13 (19.7%) studies satisfied two out of the three criteria, and only two studies (3.3%) met just one of the quality assurance criteria. The quality assessment item that was most often missing from studies was adequate reliability and validity data for the measures used, with 15.2% of studies missing this criterion. We also examined each study for the presence of validity screening (i.e., the identification and removal of mischievous or careless responders). However, only three out of the total 66 studies included some form of validity screening, thus preventing us from conducting any statistical comparison of studies.

Meta-Analytic procedure

The correlation coefficient (r) is the chosen effect size for our meta-analyses. When a correlation matrix was not provided (N  =  16), corresponding authors were emailed to request the bivariate correlations for our variables of interest. Of the remaining studies, those that included odds ratio data were transformed using the Comprehensive Meta-Analysis (CMA) software, version 3. Studies that reported a beta value from a regression analysis (N  =  7) were transformed using the method recommended by Peterson and Brown (2005). This method of transforming beta values to correlation coefficients has recently received some criticism (Roth et al., 2018). However, this solution is more favourable than excluding the studies altogether (Borenstein et al. 2011). To account for this recent critique, a moderator analysis was conducted to compare standardized data (beta) transformed to correlation coefficients and direct estimates of Pearson r extracted from studies.

All extracted correlation coefficients were transformed using Fisher's Z to approximate a normal sampling distribution (Borenstein et al., 2011). Effect sizes were weighted using the inverse variance based on the standard error from the study (Card, 2011). Both meta-analyses were conducted with the transformed data, and then effect sizes were transformed back to Pearson r coefficients to make results more readily interpretable. Given that we assumed that differences among study effect sizes were due to a combination of sampling error and other factors, we selected a random effects model. The Der Simonian and Laird estimator was used. Two meta-analyses were conducted: (a) perpetration and TSR closeness and (b) victimization and TSR closeness. Aligned with Cohen's (1992) recommendations for qualitatively assessing sample size values, effect sizes are interpreted as small (r  =  .10-.29), medium (r  =  .30-.49), or large (r > .50). CMA, version 3 was used for all computations (Borenstein et al., 2013).

Heterogeneity statistics, the Q statistic and the I² index, are reported with each of the meta-analyses. The Q statistic is a significance test that tests the assumption of homogeneity of effect sizes (Borenstein et al., 2011). A significant Q statistic indicates that the effect sizes will differ between studies and that sample and methodological characteristics (i.e., potential moderators) play a critical role in the accurate interpretation of meta-analytic results. The I² index allows for the determination of the amount of heterogeneity among the effect sizes (Card, 2011). Rough estimates of the I² index are: ∼25% is equal to small heterogeneity beyond sampling error, ∼50% is representative of moderate heterogeneity outside of sampling error, and ∼75% + is considered to be large heterogeneity independent of sampling error (Borenstein et al., 2016).

Three moderator analyses were conducted using dichotomized categorical variables to more accurately understand how the association between peer aggression involvement and TSR quality changes as a function of the level of each moderator. For the moderator analyses, a mixed-effects analysis was selected. When moderator levels included fewer than five studies, a fixed-effects model was used to combine effect sizes (Hedges & Vevea, 1998). Each moderator level included a minimum of 3 effect sizes (k ≥ 3; Borenstein et al., 2011).

Data screening

A sensitivity analysis was conducted to assess whether any one study had undue influence on the total effect sizes. The analysis indicated that no study was significantly manipulating results for the peer perpetration and peer victimization meta-analyses. No studies were identified as outliers. Despite the large range of effect sizes from the included studies, no study was identified as a true outlier, but was considered within the true range of the studies included.

Publication bias

Publication bias was assessed. Egger's linear regression (Egger et al., 1997) assesses whether or not studies were published due to statistically significant results by assessing the amount of asymmetry when studies are plotted by effect size and sample size (Card, 2011). It is assumed that there is no evidence of publication bias when p > .05. For perpetration and TSR closeness, Egger's linear regression was not significant (p  =  .69) (see Figure 2 for funnel plot). For victimization and TSR closeness, Egger's linear regression produced a non-significant result (p  =  .74) (see Figure 3 for funnel plot). Taken together, these results suggest that there are no persuasive indicators of publication bias in these meta-analyses.

OPEN IN VIEWER

Figure 2.

Publication bias funnel plot for peer perpetration and teacher-student relationship closeness. Note: CES  =  combined effect size. The imputed data point represents an adjustment of the combined effect size to account for potentially missing studies using the Trim-and-Fill method.

OPEN IN VIEWER

Figure 3.

Publication bias funnel plot for peer victimization and teacher-student relationship closeness. Note: CES  =  combined effect size.

Perpetration and TSR closeness

 Figure 4 displays the distribution of all 42 effect sizes and the summary effect (listed last in the figure) corresponding to peer aggression perpetration and TSR closeness. Effect sizes ranged from -.545 to .100. The overall aggregated effect represents a small significant negative association between perpetration and TSRs characterized by warmth and support [r  =  -.185 (95% CI. -.235 < r < -.133), Z  =  -6.886, p < .001]. The results indicate that as students are involved at increasing levels of aggressing against others, they tend to have decreasing closeness within their relationships with teachers. This supports our hypothesis that students who aggress against their peers will also experience reduced feelings of closeness, warmth, and support with their teachers. The test for homogeneity of variance was significant revealing that there was true variability between study effect sizes, beyond sampling error [Q (41)  =  1811.805, p < .001]. The magnitude of the heterogeneity outside of sampling error was very large, with an I² index approaching 100% (I² index  =  97.737).

OPEN IN VIEWER

Figure 4.

Forest plot for peer perpetration and teacher-student relationship closeness. Note: Q (41)  =  1811.805, p < .001; I²  =  97.737.

Victimization and TSR closeness

The second meta-analysis synthesized 48 effect sizes corresponding to the association between peer victimization and TSR closeness, ranging from -.460 to .120. The distribution of these effect sizes, including the summary effect, can be found in Figure 5. Results of the meta-analysis revealed a small significant negative association between victimization and TSRs characterized by closeness and warmth [r  =  -.157 (95% CI. -.190 < r < -.123), Z  =  -8.933, p < .001]. Although the effect is small in scale, these results support our hypothesis, indicating that overall, as victimization increases, closeness in TSRs decreases. The test for homogeneity of variance was significant, indicating that there was true variance among effect sizes, more than what could be attributed to sampling error [Q (47)  =  3515.449, p < .001]. The I² index  =  98.663, revealing a very large amount of between-study variance.

OPEN IN VIEWER

Figure 5.

Forest plot for peer victimization and teacher-student relationship closeness. Note: Q (47)  =  3515.449, p < .001; I²  =  98.663.

Moderator analyses

School level

The results for the school level moderator can be found in Table 3. No significant difference was found in the association between TSR closeness and perpetration (Q (1)  =  .096, p  =  .756) between students in primary school level and students in intermediate or secondary school level. The effect of school level on the association between TSR closeness and victimization was likewise non-significant (Q (1)  =  1.770, p  =  .183).

OPEN IN VIEWER

Table 3.

All moderator analysis for each meta-analysis.

Peer Aggression and Teacher-Student Relationships	School Level Category
	P/TSRclos	k	V/TSRclos	k
Elementary	-.173	15	-.127	19
Intermediate/Secondary	-.191	27	-.174	29
Q	.096		1.770
	Same Informant Category
	P/TSRclos	k	V/TSRclos	k
Self	-.202	3	-.145	8
Other	-.161	7	-.090	4
Q	2.508		5.800*
	Cross-Informant Category
	P/TSRclos	k	V/TSRclos	k
Same			-.138	12
Mixed			-.077	3
Q			7.211*
	Measure Category
	P/TSRclos	k	V/TSRclos	k
Established	-.199	28	-.176	29
Novel	-.197	6	-.114	8
Q	.001		2.228
	Data Type Category
	P/TSRclos	k	V/TSRclos	k
Correlation coefficient (un-transformed)	-.206	36	−156	40
Beta (transformed)	-.061	6	-.158	8
Q	4.208*		.001

Note: P = Perpetration, V = Victimization, TSRclos = Teacher-student relationship closeness. * indicates p-value <.05.

Moderator analyses

To provide a more nuanced understanding of the correlation between peer aggression involvement and TSR closeness, three moderators were investigated, the first of these being school level. The results of the analysis revealed that school level (dichotomized into elementary; k-6 and intermediate/secondary; 7–12) did not significantly moderate the relationship between peer aggression involvement and TSR closeness (see Table 3). It appears that despite clear changes in developmental stage and school context, the association between peer aggression involvement and TSR closeness remains stable. That is, the links between involvement in peer aggression and TSR closeness remain just as important in grade 12 as they do in kindergarten, and all the grades in between. One explanation for this non-significant result is that while the correlation coefficients remain stable, the individual data points may still change. It is entirely possible that two things are happening simultaneously; (1) that both peer aggression involvement rates and TSR closeness ratings are lower for older student compared to younger student populations in our sample of studies – which would align with extant literature (Kennedy, 2021; Lynch & Cicchetti, 1997; Wu & Hughes, 2015), and (2) that peer aggression involvement and TSR closeness ratings are decreasing in parallel. Thus, despite our initial hypothesis, the correlation coefficients for older and younger students would share the same amount of variability, resulting in no statistically meaningful differences. We are not able to confirm this explanation, given the limited availability of individual data points for peer aggression involvement and TSR closeness in our set of studies. It would be informative for future research to explore the association between peer aggression involvement and TSR closeness across student age and school context more deeply.

The informant moderator analyses indicated that informant type (self versus observer, observer including peers or teachers for the purposes of this study) significantly influenced results for one of the two meta-analyses. First, self-reports of peer victimization and TSR closeness yielded stronger effect sizes relative to observer reports. Studies have found that self-reports of involvement in aggressing against others and being victimized share limited agreement with reports from teachers and peers (Cornell & Brockenbrough, 2004). Second, a significant moderating effect of cross informant was found for peer victimization and TSR closeness with larger associations for studies that used the same informant for both involvement in peer aggression and TSR closeness compared to studies that employed a cross-informant approach. This analysis provides quantitative evidence that using the same informant for measures of both peer victimization specifically and TSR closeness may inflate effect sizes. Our results from this cross-informant moderator analysis are aligned with existing literature supporting the use of multiple informants (e.g., Achenbach et al., 1987). Bolstered by these findings and given the discrete differences among informants when assessing both peer aggression involvement and TSR closeness, we suggest that several varied informant perspectives be sought so a more comprehensive understanding can be gathered (Shakoor et al., 2011). When the same informant reports on different constructs on similarly structured scales, biases in the informant can influence responses to both measures in similar ways and cause shared method variance. This can be true of self and other (teacher and peer) informants. Soliciting various informant perspectives removes this methodological bias and potentially provides a more accurate estimate of associations between the two constructs. Importantly, our moderator analyses only included primary school aged children since no studies in our sample that included intermediate/secondary school aged children utilized observer-reports for measures of peer aggression involvement. It would be worthwhile for future research to adopt observer-report measures when investigating involvement of peer aggression and TSR quality in older age groups to further assess the function of age in the association between these two constructs.

While many studies included in our analyses used measures with strong validity evidence (e.g., STRS; Pianta, 2001), several other studies included measures that had been specifically designed for the purposes of the single study, or included a measure that combined items from several different measures. This moderator analysis of measure type suggests that using novel measures does not necessarily result in smaller effect sizes. Thus, attempts to use novel measures should not be entirely dissuaded, particularly if it allows researchers to administer a measure that better suits their study requirements. However, what remains most important, and especially so when utilizing novel measures, is that researchers provide psychometric properties of the included measures whenever possible, to ensure that their chosen measure is indeed assessing the desired construct. The consolidation and validation of measures in these domains will also make future quantitative syntheses more feasible and accurate.

Limitations

An important limitation requiring consideration is the significant heterogeneity of the studies included. Notably, this was an important strength of the paper as it allows the results to be generalized to diverse populations across several different countries and continents. However, the large amount of heterogeneity complicates efforts to explain how and why exactly effect sizes differed from one study to the next. It will be important for future research to explicate the subtleties of the relationship between the two constructs. Additionally, future analyses should assess the reduction of heterogeneity after accounting for particular moderators. While these analyses were ultimately beyond the scope of our present meta-analyses, a closer investigation of the impact of moderators on the link between peer aggression involvement and TSR closeness is recommended to better inform prevention and intervention programs for a variety of students.

While our set of studies was diverse, Africa was not represented in the present analysis and the majority of the included studies were North American, and primarily from the United States. This highlights the need for more geographically diverse research to test the generalizability of our findings to children, youth, and teachers in other parts of the globe.

Additionally, given the cross-sectional nature of our analyses, we are not able to make any conclusions regarding the causal ordering of the variables. It would be worthwhile for future research to focus on quantitatively synthesizing the longitudinal associations between peer aggression involvement and TSR closeness to better understand how the two constructs interact across time.