It Takes a Village to Raise a Child:
A Multilevel Meta-analysis of the Association between Natural Mentoring and
Youth Outcomes
Masterscriptie Forensische Orthopedagogiek
Faculteit der Maatschappij- en Gedragswetenschappen
Universiteit van Amsterdam
Bo Wildschut en Dafne Smit 11252235 – 10845658 Begeleiders: Geert-Jan Stams, Levi van Dam, Susan Branje & Mark Assink Amsterdam, 17 juli 2017
2
Abstract
This study is a systematic review and multilevel meta-analysis of the association between adolescents’ involvement in natural occurring mentoring relationships, the quality of these relationships, and several outcome domains. The review included 30 studies on the topic dating from 1992 to present. Two separate meta-analyses were conducted on the relation between the presence of a natural mentor, and the quality of natural mentoring relationships on youth outcomes. As a guiding conceptual framework for our analysis, a theoretical model of youth development and the potential of natural mentors is drawn, indicating that these natural mentoring relationships foster positive youth development and buffer against the risks associated with the tumultuous years of adolescence. The findings indicate that the presence of a natural mentor is significantly associated with positive youth outcomes (r = .097). Secondly, when taking quality into account a somewhat larger effect size was found (r = .160). Several moderators were included in the study, which included participant characteristics, mentor/relationship characteristics and study characteristics. Stronger relations were found for the academic and social-emotional domain and having a natural mentor in a professional role. Implications for theory and practice concerning the quality of natural mentoring relationships are discussed.
3
Introduction
The rearing environment of a child does not only consist of its parents. The extended family network, friends, neighbors and teachers also play a role or take responsibility for the education, development and well-being of the child (Bowers, Johnson, Warren, Tirrell, & Lerner, 2015; Kesselring, De Winter, Van Yperen, & Lecluijze, 2016). This is reflected in the statement ’it takes a village to raise a child’, which stands for the educational civil society. The importance of this broader social network increases during adolescence, when adolescents are biologically, emotionally, and developmentally wired for engagement beyond their families, and increasingly gain psychological and behavioral autonomy from their parents (Bowers et al., 2014; Fruiht & Wray-Lake, 2013; Patton et al., 2016).
If relationships are supportive, they may develop into natural mentoring relationships that foster positive youth development and buffer against the risks associated with the tumultuous years of adolescence (Bowers et al., 2015). Since research on the effects of natural mentoring during adolescence is steadily growing, and results have not been consistent across outcomes or in some instances were even equivocal (DuBois & Silverthorn, 2005a; Rhodes, Contreras, & Mangelsdorf, 1994; Zimmerman, Bingenheimer, & Notaro, 2002), a review of the literature seems timely to integrate the current knowledge on natural mentoring and explain differences within and between studies. The present study is a systematic review and multilevel meta-analysis of the association between adolescents’ involvement in natural occurring mentoring relationships, the quality of these relationships, and social-emotional, health (psychological and physical), academic and vocational outcomes.
Natural mentoring relationships develop in youth’s everyday context, and usually arise from already existing relationships. A natural mentor may be a relative, neighbor, teacher, friend or someone from a religious community other than the parents or step-parents, who is a confidant and spokesman for the youth (Schwartz, Rhodes, Spencer, & Grossman, 2013;
4 Spencer, Tugenberg, Ocean, Schwartz, & Rhodes, 2016; Van Dam et al., in press). Natural mentors may enhance the strengths of youth during their identity development, in their ability to form a social (e.g., interpersonal, role, or group) identity through one’s sense of belonging and mattering with significant others (Bowers et al., 2012; Erikson, 1968; Lerner, Von Eye, Lerner, & Lewin-Bazin, 2009). Also, by being a companion for youths and providing reliable support, natural mentors may establish and maintain resilience (Southwick, Morgan, Vythilingam, & Charney, 2007).
The potential influence that natural mentors may have, ranging from promotive to protective, depends on the level of adversity and number of risk and protective factors youths have (Hurd & Sellers, 2013). For normally developing youths, natural mentors may play a significant role in helping youths cope with difficulties, achieve goals, and navigate their identity. For youths at-risk, the natural mentoring relationship has the potential to offset individual and contextual risks, with adolescents often attributing their capacity to thrive despite adversity to the support of a caring adult (Greeson & Bowen, 2008). These close personal relationships may promote feelings of predictability and stability, and enhance well-being in the life of all youths (Cohen & Wills, 1985).
The present meta-analysis is guided by the theory of Positive Youth Development [PYD], which describes the normative developmental tasks of typically developing youth dealing with general day-to-day stressors or mild adversity (Damon, 2004), and resilience theory, which explains the mechanisms that buffer against risks when adolescents are faced with more severe adversity or disturbed development (Ungar, 2011). Figure 1 shows a continuum from typical (normative) to atypical (disturbed) youth development and the potential influence of a natural mentor.
5 Figure 1: Theoretical model of youth development and the potential of natural mentors.
For typically developing youths, natural mentors predominantly serve as promotive resources by providing information, cognitive guidance, advice and material or instrumental support, which can be useful in healthy decision making, coping with day-to-day stressors, and job search (Erickson, McDonald, & Elder, 2009). When identity development becomes central, mentor guidance may help shift youths’ conceptions of both their current and future vocational identities (Erikson, 1968). Also, through social interactions with natural mentors, adolescents acquire and refine new cognitive skills, and become more receptive to adult instruction and perspectives (Radziszewska & Rogoff, 1991). These interactions influence the adolescents’ brain development and shape the capabilities youths take forward into adult life (Blakemore & Mills, 2014). Last, natural mentors may serve as role models or open doors to activities, resources and educational or occupational opportunities that youth can draw on to construct their sense of identity (Darling, Hamilton, Toyokawa, & Matsuda, 2002). This is referred to as ‘possible selves’: youths’ ideas of what they might become or what they would like to become in the future (Markus & Nurius, 1986; Schwartz et al., 2013). For these
6 typically developing youths, the relationship with natural mentors may focus more on the achievement of skills and the improvement of critical thinking than on emotional issues (Rhodes, Spencer, Keller, Liang, & Noam, 2006).
When stressors or adversities accumulate, youth may turn to their natural mentors for social support as an adaptive form of coping and resilience (Hurd & Zimmerman, 2010a). Resilience theory assumes that in the presence and experience of adversity (i.e., low SES, learning disabilities, low grades), resilience derives from both the capacity of the individual and the capacity of his or her social and physical ecologies to facilitate coping in meaningful ways (Ungar, 2011). Natural mentors can accomplish a compensatory role for youths who experience adversity (Southwick et al., 2007). By modeling, caring, and providing support, natural mentors can both challenge negative views that some youth may hold of themselves and demonstrate that positive relationships with adults are possible. The natural mentoring relationship may become a ‘corrective’ experience for youth who have experienced unsatisfactory relationships with parents or other caregivers (Rhodes, 2005). In this way, natural mentors may counteract or neutralize the effects of risks that youths face (Zimmerman et al., 2002).
For youth who grow up facing the challenges of severe and ongoing adversity in their families and communities (i.e., youth in foster care, adolescent mothers), the availability and social support of at least one caring adult (i.e., natural mentor) predominantly plays a protective role (Smith & Carlson, 1997). As a protective source of support, natural mentors modify the relation between risks and outcomes by lessening the effect of risk factors and enhancing the effects of existing compensating factors. Natural mentors may function as advocates or secondary attachment figures, as at-risk youths develop more intense bonds with their natural mentors to satisfy their emotional and social needs (Bowlby, 1988; Erdem, DuBois, Larose, De Wit, & Lipman, 2016; Rhodes et al., 2006).
7 Relationships with natural mentors are thus normative for youth in the general population as well as protective for marginalized youth who are at-risk due to adverse individual or environmental circumstances (Thompson, Greeson, & Brunsink, 2016; Stevenson & Zimmerman, 2005). Indeed, several meta-analyses indicate a positive association between youth mentoring and improved psychosocial, behavioral, academic, and vocational outcomes (DuBois, Portillo, Rhodes, Silverthorn, & Valentine, 2011; Eby et al., 2013). However, these studies did not focus on only natural mentoring relationships, but also on matched relationships between volunteer mentors and mentees. A systematic literature review is the most advanced research method used on the topic of natural mentorship. The review by Thompson et al. (2016) focused on the earlier described outcome measures. This review reported a positive association between the presence of a natural mentor and positive well-being outcomes among foster youth. General conclusions for the youth population could not be made based on this review, since it only focused on youth in foster care. To compare the results from the present study to the results of this review, in order to draw general conclusions, it is necessary to conduct research on the same outcome variables.
The association between natural mentoring relationships and youth outcomes may be affected by a range of individual, relational, and contextual moderators, whereby the quality of natural mentoring relationships appears to be a particularly important determinant of youth outcomes (Goldner & Mayseless, 2009; Grossman & Rhodes, 2002; Grossman, Chan, Schwartz, & Rhodes, 2012; Parra, DuBois, Neville, Pugh-Lilly, & Povinelli, 2002). Key features of the relationship quality are a mentor’s role, emotional closeness, frequency of contact and relationship duration (Rhodes, 2002). For example, a greater frequency of contact and a longer lasting relationship create opportunities for more involvement and closeness between the mentor and youth (Whitney, Hendricker, & Offutt, 2011). Consequently, youths will feel more supported when there is a close emotional bond, and might be more likely to
8 take advice from their natural mentor (Hurd & Sellers, 2013). The frequency of contact and length of the relationship are possibly essential for whether processes of change have the opportunity to occur in the relationship (Karcher, Nakkula & Harris, 2005; Spencer et al., 2016). Last, the demographic background of the mentor also tends to be important for the quality of the mentoring relationship. Being more familiar with the youth’s cultural and personal background (e.g., same ethnicity, same gender), natural mentors are likely to have better perceptions of the support needed and may provide more impactful and appropriate guidance (Whitney et al., 2011).
Other factors may also moderate the effects of natural mentoring relationships. Youths who are overwhelmed by social and behavioral problems may lack strong, enduring ties with their mentors and, perhaps consequently, experience relatively fewer benefits. In fact, risk factors are more numerous and stronger than promotive and protective factors because many risk factors are related, cumulate in negative outcomes, are relatively stable across time and context and thus prevent the development of promotive and protective factors (Vanderbilt-Adriance & Shaw, 2008). Youth who only encounter day-to-day stressors or mild adversity may not need much support from their natural mentors to have positive outcomes, since they have fewer stressors to deal with (Rhodes et al., 2006).
Furthermore, natural mentors can be segregated into a variety of roles in the life of youths. Family (kin) members serve as mentors more often for younger adolescents, whereas non-familial (non-kin) and professionals (i.e., teachers) mentors most commonly develop during secondary school (Fruiht & Wray-Lake, 2013). This is consistent with their developmental stage, when youths build identities outside the family and autonomy from their parents increases dramatically (Bowers et al., 2014; Hurd, Stoddard, Bauermeister, & Zimmerman, 2014). Additionally, lower-income youth tend to nominate more kin support, while higher-income youths are more likely to nominate teachers, counselors, and other
non-9 kin mentors. Although kin relationship ties tend to be more intensive, they may be less able to serve as ‘bridging’ capital that can link youth to new educational and occupational opportunities (Raposa, Erikson, Hagler, & Rhodes, in press). Kin relationships might provide more emotional support, whereas non-kin mentors might provide more informational or instrumental support, which can help achieving academic or career goals (DuBois & Silverthorn, 2005a).
Finally, although typically unexamined in the mentoring literature, major factors that have been consistently shown to predict impacts in meta-analyses from other fields are study characteristics, such as the methodological rigor and year of study (Cheung & Slavin, 2016; Saha, Saint, & Christakis, 2003). As the field of natural mentoring has matured, and researchers have converged on validated measures and representative samples, the rigor of published research has improved. Published studies in higher rated journals may report stronger impacts than unpublished reports due to biases in publishing only the significant results (Cheung & Slavin, 2016).
The present multilevel meta-analytic review examines the relation between natural mentoring and outcomes in various domains of adolescent functioning, accounting for both within and between study differences in effect sizes. Variables which have been shown to be significant moderators in previous studies on informal mentoring as well as potential moderating variables that were neglected in past studies are examined to test which individual, relational and study factors moderate the association between natural mentoring relationships and youth outcomes. Particular attention is given to the role of mentoring relationship quality in shaping youth outcomes.
Several (exploratory) hypotheses are proposed. The first hypothesis states that the presence of a natural mentor compared to no mentor has a positive effect on youth outcomes. Second, quality of natural mentoring relationships is expected to be positively associated with
10 youth outcomes. Third, the involvement of non-familial mentors and professional mentors is expected to yield larger associations than involvement of kin mentors. Finally, the putative moderating influence of risk status is tested as an explorative hypothesis. A significant effect of risk status would support the fruitfulness of our theoretic model, providing preliminary empirical evidence that the function of mentoring may be different for various levels of environmental adversity affecting adolescent development.
Method Sample of Studies
All studies addressing the relation between natural mentoring relationships and youth outcomes published before May 2017 were included in the current meta-analytic review. Articles published in scientific journals, books, and unpublished reports were found in the following databases: ERIC, PsychINFO, PubMed, Wiley Online Library, and Google Scholar. The search string included two or three elements: a mentor element, an outcome element and if needed an age element. For the mentor element, the following definitions of natural mentor were used: ‘natural mentor*’, ‘informal mentor*’, ‘youth mentor*’, ‘important non-parental adult*’, ‘naturally acquired mentoring relationship*’, ‘mentoring adolescen*’, ‘VIP’, or ’YIM’. For the outcome element, the following keywords were used: ‘youth outcom*’, ‘behavior outcom*’, ‘academic outcom*’, ‘foster care’, ‘youth care’, ‘delinquency’, ‘internalizing problem*’, ‘externalizing problem*’, ‘psychopathology’, ‘social-emotional’, and ‘work-related outcom*’. For the age element, the keywords youth* or adolescen* were used. Additionally, reference lists of the usable articles were inspected to find additional relevant studies. Not all studies that came across could be included, as some could not be traced in any digital library. Further, authors were contacted to retrieve relevant studies and missing study information as much as possible.
11 Multiple inclusion criteria were formulated to select the studies for this meta-analysis. First, youth outcomes had to be operationalized as social-emotional, physical health, psychopathology, academic, and/or vocational outcomes. Second, the mean age of the sample had to be between age 10 and 24. Third, the natural mentor had to be a different person than the youth’s parents or step-parents and fulfilling the role of an important person in the life of the youth. Fourth, the connection between the mentor and youth had to be an already existing relationship. Studies measuring effects of mentoring programs with natural mentors were excluded. A common problem in performing a meta-analysis is that studies may not have been published because of nonsignificant or unfavorable findings, the so called ‘publication or file drawer bias’ (Rosenthal, 1995). Therefore, it is possible that the studies included in the meta-analysis are not an adequate representation of all previously conducted studies on this topic. To reduce the problem of publication bias in our results, unpublished studies were screened by searching the ResearchGate database and several authors were contacted and asked for unpublished studies. Finally, the full publication lists of well-known authors (i.e., DuBois, Hurd, Rhodes, Zimmerman) in the field of natural mentoring were screened for additional studies that could not be found in the databases.
The three first authors conducted the screening and selection process. When in doubt, the last authors were consulted. Appendix 1 presents a flow chart of the search. The initial search resulted in 281 articles, which also contained reviews and qualitative studies. This was narrowed down to 33 articles by inspection of the title and abstract. By using the ancestry method on these 33 articles, 39 new articles were included. By thoroughly examining full texts of the 72 studies, 42 studies were excluded because they did not fit the inclusion criteria. A total of 30 studies (with 222 effect sizes) met the inclusion criteria. Table 1 provides an overview of the included studies and their characteristics. Included studies in the present
12
Table 1
Characteristics of Included Studies
Author (year) N Peer review
IF Design Informant Continent Type of outcome Sex Ethnic minority Mean age Sample type Presence natural mentor DuBois & Silverthorn (2005a)
3187 Yes 4.14 Long Self USA Mixed B/G 32.60 21.40 General
population (low-risk) Hurd & Sellers
(2013)
259 Yes 1.56 Cross Self/ Teacher
USA Mixed B/G 100 13.56 At risk
population (medium-risk) Zimmerman, Bingenheimer, & Notaro (2002)
770 Yes 2.07 Cross Self USA Mixed B/G 82.80 17.50 At risk
population (medium-risk) Ahrens, DuBois, Lozano, & Richardson (2010)
1714 Yes 1.22 Cross Self USA Mixed B/G 25 16 At risk
population (high-risk) Ahrens, DuBois,
Richardson, Fan, & Lozano (2008)
310 Yes 5.47 Cross Self USA Mixed B/G 35 16 At risk
population (high-risk) Rhodes, Ebert,
& Fischer (1992)
129 Yes 2.15 Cross Self USA Psychopat
hology
Girls 100 18.07 At risk population (high-risk)
13 Contreras, Mangelsdorf, (1994) population (high-risk) Collins, Spencer, & Ward (2010)
96 Yes .38 Cross Interviewer USA Mixed B/G 47 19 At risk
population (high-risk) Dang, Conger,
Breslau, & Miller (2014)
197 Yes 1.28 Cross Self USA Mixed B/G 75.6 18 At risk
population (high-risk) Erickson,
McDonald, & Elder (2009)
12621 Yes 2.86 Long Official registration
USA Academic B/G 35 21.72 General population (low-risk) Hurd & Zimmerman (2010a) 615 Yes 2.15 Cross/ Long
Self USA Psychopat
hology B/G 100 17.51 At risk population (medium-risk) Hurd & Zimmerman (2010b) 93 Yes 2.48 Cross/ Long
Self USA Psychopat
hology Girls 100 17.66 At risk population (high-risk) Hurd, Stoddard, Bauermeister, & Zimmerman (2014)
3334 Yes 2.05 Cross Self USA Mixed B/G 24.6 20.8 General
population (low-risk) Hurd, Varner, &
Rowley (2013)
259 Yes 3.56 Cross Self USA
Socio-emotional B/G 100 13.56 General population (low-risk) Sánchez, Esparza, & Cólon (2008)
140 Yes .80 Cross Self/ official registration
USA Academic B/G 100 17.88 At risk population
(medium- risk) Cavell, Meehan,
Heffer, &
95 Yes 1.33 Cross Self USA Psychopat
hology
B/G 17 18.7 At risk
14
Holladay (2002) (high-risk)
Kogan, Brody, & Chen (2011)
375 Yes 2.15 Long Self/ composite
USA Mixed B/G 100 17 At risk
population (medium- risk) Greeson, Usher, & Grinstein-Weiss (2010)
7977 Yes .97 Long Self USA Vocational B/G 20 21.28 General
population (low-risk) Hagler, Raposa,
& Rhodes (in press)
193 - - Long Self USA Mixed B/G 43.5 11.2 At risk
population (medium-
risk) McDonald &
Lambert (2014)
16386 Yes 2.15 Long Self USA Vocational B/G - 22.5 General
population (low-risk)
Linnehan (2003) 47 Yes 2.76 Long Self USA Mixed B/G 79 17.45 At risk
population (medium- risk) McDonald, Erickson, Johnson, & Elder (2007)
5740 Yes 1.77 Long Self USA Vocational B/G 35 - General
population (low-risk) Erickson &
Philllips (2012)
8379 Yes 1.23 Long Self USA Academic B/G 47 15.36 General
population (low-risk) Munson &
McMillen (2009)
339 Yes .97 Long Self USA Mixed B/G 55 19.04 At risk
population (high-risk) Relationship
15 Not e. N = nu mb er of part icip ants ; pee r revi ew = pub lish ed in pee r revi ewe d arti cle (yes /no) ; IF = impact factor of journal; design = cross-sectional (cross) or longitudinal (long); informant = self-report (Self), interviewer-report (Interviewer) or Hurd, Varner, &
Rowley (2013)
259 Yes 3.56 Cross Self USA
Socio-emotional B/G 100 13.56 General population (low-risk) Chang, Greenberger, Chen, Heckhausen, & Farruggia (2010)
754 Yes 2.48 Long Self USA Mixed B/G 77 17.50 General
population (low-risk)
Schwartz, Chan, Rhodes, & Scales (2013)
1860 Yes 1.97 Cross Self USA
Socio-emotional B/G 44.4 15 General population (low-risk) Sánchez, Esparza, & Cólon (2008)
140 Yes .80 Cross Self/ official registration
USA Academic B/G 100 17.88 At risk population
(medium- risk) Kogan, Brody,
& Chen (2011)
116 Yes 1.79 Cross Self/ official registration
USA Mixed B/G 100 19.5 At risk
population (medium-
risk) Bowers et al.
(2012)
710 Yes 1.97 Long Self USA
Socio-emotional Boys 21.1 15.77 General population (low-risk) Black, Grenard, Sussman, & Rohrbach (2010)
3320 Yes 1.67 Long Self USA Psychopat
hology
B/G 59 15.3 General population
(low-risk) Klaw & Rhodes
(1995)
204 Yes 2.40 Cross Self USA Mixed Girls 100 15.9 At risk
population (high-risk)
16 teacher-report (Teacher); type of outcome = internalizing, overall and/ or externalizing (psychopathology), social confidence and/ or confidence (social-emotional), academic, vocational or two or more different outcome domains (Mixed); sex = only girls (Girls), only boys (Boys) or boys and girls (B/G); ethnic minority = proportion of non-Caucasian.
17
Coding Studies and Potential Moderators
The three first authors of this article coded the included studies according to the suggestions of Lipsey and Wilson (2001). The outcome variables in these meta-analyses were youth outcomes in several life domains. The predicter variables were the presence of a natural mentor or the quality of the natural mentoring relationship. Five studies (#ES = 36) were double coded by two of the first authors. It is common to calculate the inter-rater agreement, which proved to be good with 94% agreement between the two coders.
Each study was coded on multiple characteristics. The characteristics could be divided into six major categories: report information (year of report, impact factor, published/unpublished); evaluation methodology (study design, uni-/multivariate, sample size, criteria for natural mentor inclusion, type of reporter); characteristics of participating youth (gender, ethnicity, age, school type, at-risk status); mentor–mentee relationships (mentor demographics, gender matched, ethnicity matched, type of mentor, length of relationship, kind of support, frequency of contact, feelings of closeness); and assessment of outcomes (type of domain, sort of comparison, type of association, type of measure, reliability).
The following report information and evaluation methodology characteristics were examined as moderators in the meta-analysis. First, the impact factor of the journal in which the study was published (continuous variable) was coded, because the impact factor is a first indication of study quality (Saha et al., 2003). Second, the year of publication (continuous variable) was coded. It was expected that the quality of older studies was lower than the quality of more recent studies, as the statistical and methodological knowledge has increased largely in social science research over the last decades. Finally, the study design was coded (cross-sectional vs. longitudinal design), as cross-sectional studies measure the relation between natural mentoring and youth outcomes at one point in time, and longitudinal studies
18 can take the developmental aspect of the relation between natural mentoring and youth outcomes into account.
The following sample characteristics were coded: mean age, proportion of males (continuous variable), and the proportion of youth with a minority background (non-Caucasian) in the sample (continuous variable). The capacity and willingness of youths to forge close connections with natural mentors may vary as a function of their age. At-risk status was coded, as effects of natural mentors may differ for at-risk youths and normally developing youths (DuBois & Silverthorn 2005a; Werner, 1993). Youths were coded at-risk when it was (explicitly) stated that the used sample was an at-risk population. Risk status was divided into three levels: low-, medium- or high-risk. Youths from the Adolescent Health sample or general youth samples were coded as low-risk. The following risk populations were coded as medium-risk: youth with learning disabilities; youth with GPA < 3; youth on the Big Brothers, Big Sisters waiting list; rural African American young males; economically disadvantaged youths. At last, African American adolescent mothers, young Latino mothers, homeless youth, youth in foster care, youth of alcoholic parents, and pregnant and parenting teenagers were coded as high-risk.
Several mentor-mentee relationships characteristics were coded, because available evidence points to more positive outcomes when youths experience longer and more connected relationships with their mentors (DuBois & Silverthorn, 2005b; Hurd & Sellers, 2013; Hurd, Varner, & Rowley, 2013; Karcher et al., 2005). It was coded whether youth and mentors had the same gender and/or ethnicity (continuous variable). Type of mentor was coded into three categories: kin (e.g., grandmother, grandfather, aunt, uncle, older sibling); non-kin (e.g., sport coach, employer, co-worker, neighbor, friend, friend’s parents); and professional (e.g., teacher, guidance counselor, minister/priest/rabbi, religious leader, doctor/ therapist). The predominantly reported frequency of contact (daily or weekly) and feelings of
19 closeness (very close or quite close) were coded. Finally, the average length of relationship was coded in years (continuous variable) and the percentage of informational, instrumental and emotional support youths receive from their natural mentor (continuous variable).
Similar to other published meta-analyses and studies on youth mentoring, outcome variables that were conceptually similar were combined (DuBois & Silverthorn, 2005b; DuBois et al., 2011; Eby, Allen, Evans, Ng, DuBois, 2008; Eby et al., 2013). This was necessary to draw general conclusions about the relation between mentoring and youth outcomes. Table 2 lists the five broad categories of outcomes. For each category, the specific outcomes were examined and operationalized. For the psychopathology domain, four categories of outcomes were used for the analysis on the presence of a natural mentor, since these categories represent a particular type of behaviour. This distinction could not be used for the meta-analysis on quality of the natural mentoring relationship, since there were not enough effect sizes for these particular categories. The same procedure was used for the social-emotional domain. Only for the meta-analysis on the presence of a natural mentor a distinction between social competence and confidence could be made
20 Table 2 Op era tion aliz ing of Out co me Do mai ns, incl udi ng Examples of Assessed Variables in Each Domain
Domain
Academic High school completion, school attendance, academic engagement, higher grades, absences Social-emotional
Social competence Social skills, prosocial behavior, negative life events, self-regulation
Social confidence Self-esteem, life satisfaction, well-being, perceived social support, care, character, connection
Vocational Economic benefits, fulltime employment, job autonomy, discontinuous employment, repetitive work, Supervisory authority, school importance, school belonging
Physical health General health, physical activity, birth control, condom use, Body Mass Index above 25, Sexually Transmitted Disease Diagnosis
Psychopathology
Internalizing Depression, anxiety, suicidal ideation, psychosomatic symptoms, mental health
Conduct problems Substance use, sexual risk behavior, delinquency, problem behavior, aggression, rule breaking Overall psychopathology Global severity, SCL-90-R
21
Calculation of Effect Sizes and Analyses
Two separate meta-analyses were conducted to assess the overall relation between the presence of a natural mentor and youth outcomes. The first meta-analysis focused on the presence or absence of a natural mentor, in which most studies could be included. Since some studies focused on the quality and availability of a natural mentor (e.g., Likert-scale of availability, less or more connected mentor), instead of the presence or absence, the second meta-analysis was conducted to assess the relation between the quality of the natural mentoring relationships and youth outcomes.
Effect sizes were transformed into the correlation coefficient r. For the first meta-analysis, a positive correlation indicates that the presence of a natural mentor relationship is associated with positive youth outcomes, whereas a negative correlation means that the presence of a natural mentor relationship is associated with negative youth outcomes. For the second meta-analysis, a positive correlation indicates that a higher quality of the natural mentoring relationship is associated with positive youth outcomes, whereas a negative correlation indicates that a lower quality of the natural mentoring relationship is associated with negative youth outcomes.
Effect sizes were calculated using the calculator of Wilson (2013) and formulas from Lipsey and Wilson (2001). If an article only mentioned that the relation was not significant, an effect size was coded as zero (Lipsey & Wilson, 2001). Continuous variables were centered around the mean, and categorical variables were recoded into dummy variables. Correlation coefficients r were recoded into Fisher z-values (Lipsey & Wilson, 2001). After the analyses, the Fisher z-values were transformed back into correlation coefficients for interpretation and reporting purposes. Standard errors and sampling variances of the effect sizes were estimated using formulas by Lipsey and Wilson (2001).
22 By including multiple effect sizes per study, the assumption of independent effect sizes that underlie classical meta-analytic strategies was violated (Hox, Moerbeek, & van de Schoot, 2010; Lipsey & Wilson, 2001). To deal with the interdependency of effect sizes, a multilevel random effects model was applied (Houben, Van Den Noortgate, & Kuppens, 2015; Van Den Bussche, Van Den Noortgate, & Reynvoet, 2009; Viechtbauer, 2010). This model is adequate and often used for multilevel meta-analyses, and in general superior to the fixed-effects approaches used in traditional meta-analyses (Van Den Noortgate & Onghena, 2003). A multilevel approach has the advantage that it accounts for the hierarchical structure of the data by nesting effect sizes within studies. In this way, multiple effect sizes can be extracted from each included primary study, so that all information in the studies can be preserved and maximum statistical power is achieved (Assink & Wibbelink, 2016). A three-level random effects model was used to account for three three-levels of variance, including the sampling variance of each effect size (level 1), the variance between effect sizes extracted from the same study (level 2), and the variance between the studies (level 3). The meta-analyses were conducted in R (version 3.4.0) with the metafor-package, using the syntax from Wibbelink & Assink (2016).
To estimate the model parameters, the restricted maximum likelihood estimate (REML) was applied (Van Den Noortgate & Onghena, 2003). The t-distribution was used for testing individual regression coefficients of the meta-analytic models and for calculating the corresponding confidence intervals (Knapp & Hartung, 2003). The Knapp and Hartung method (2003) has the advantage that it reduces Type I-errors (Assink & Wibbelink, 2016). When models were extended with categorical moderators consisting of three or more categories, the omnibus test of the null hypothesis that all group mean effect sizes are equal, followed an F-distribution. Likelihood ratio tests were used to compare the deviance scores of the full model to the deviance of models excluding level 2 or level 3 variance parameters,
23 making it possible to determine whether significant variance is present at the two levels (Assink & Wibbelink, 2016). In case there was significant variance on either of these two levels, the distribution of effect sizes was considered to be heterogeneous. This indicates that the effect sizes could not be treated as estimates of a common effect size, and moderator analyses were performed.
Although several efforts were made to reduce publication bias in the search strategy, this could not guarantee the absence of publication bias or other forms of bias in the results. To assess the influence of publication bias, a funnel plot asymmetry according to Egger’s method was tested first (Egger, Smith, Schneider, & Minder, 1997). A funnel plot is a scatter plot of the effect sizes against the effect sizes’ precision (the inverse of the standard error). In case of publication bias, a gap in the effect size distribution would be present, showing an asymmetrical funnel plot and a significant Egger’s test. Second, a trim and fill procedure was performed, by drawing a trim and fill plot in MIX 2.0 (Bax, 2011; Duval & Tweedie, 2000). The trim and fill procedure corrects for funnel plot asymmetry by imputing estimated missing effect sizes that are calculated on the basis of existing effect sizes. If the trim and fill plot showed missing effect sizes, estimated effect sizes of missing studies were imputed, and the multilevel meta-analyses were reran in R, as this shows the influence of the estimated missing data on the overall effect of the meta-analyses. Finally, the skewness of the effect size distribution was calculated in SPSS, because if publication bias is present, a skew distribution of the effect sizes would be expected (Begg & Mazumdar, 1994).
Results
The results of the two meta-analyses are described below. Table 3 shows the overall relation between the presence of a natural mentor and youth outcomes, and the overall relation between the quality of natural mentoring relationships and youth outcomes.
24
Table 3
Overall Relation between the Presence of a Natural Mentor and the Quality of Natural Mentoring Relationships on Youth Outcomes
Outcome k #ES Mean
r
95% CI p σ2level 2 σ2level 3 % Var. Level 1 % Var. Level 2 % Var. Level 3 Presence natural mentor Youth- outcomes 23 166 .097 .056; .139 <.001 *** 0.022*** 0.006*** 1,00 79,00 20,00 Relationship quality Youth- outcomes 8 56 .160 .052; .268 .004** 0.013*** 0.019*** 2,00 41,00 57,00 Note. Youth outcomes = academic, social-emotional, vocational, physical health,
psychopathology; k = number of studies; #ES = number of effect sizes; mean r = mean effect size (r); CI = confidence interval; σ2level 2 = variance between effect sizes extracted from the same study; σ2
level 3 = variance between studies; % Var = percentage of variance distributed.
* p ≤ .05. ** p ≤ .01. *** p ≤ .001.
3.1. Overall relation between the presence of a natural mentor and youth outcomes
The meta-analysis on the relation between the presence of a natural mentor and youth outcomes contains 24 independent studies (k), reporting on 166 effect sizes (#ES), and a total sample of N = 63.327 participants. A small, significant relation was found between the presence of a natural mentor and youth outcomes (r = .097; 95% CI = .056; .139; p <.001), indicating that the presence of a natural mentor is significantly associated with more positive youth outcomes.
When checking for publication bias, Egger’s test was significant (t = 7.712, p < .001), indicating that there was funnel plot asymmetry (Egger et al., 1997). Next, a trim and fill procedure was conducted. Appendix 2 shows 46 missing effect sizes on the right side of the funnel plot. When taking these 46 missing effect sizes into consideration, the overall effect was r = .193 instead of r = .097. These results suggest a substantial underestimation of the true effect.
25 The results of the likelihood-ratio tests showed that there was significant variance between effect sizes from the same study (i.e., level 2 variance) and that there was significant variance between studies (i.e., level 3 variance). Since the variances at level two and three were significant, it was concluded that there was heterogeneity among the effect sizes that may be explained by characteristics of natural mentoring relationships, studies and samples. Therefore, moderator analyses were conducted.
3.2. Moderator analyses on the relation between the presence of a natural mentor and youth
outcomes
Table 4 presents the moderator analysis on the relation between the presence of a natural mentor and youth outcomes. The type of outcome domain (academic; social competence; social confidence; vocational; physical health; internalizing; conduct problems; substance use, overall psychopathology) had a moderating effect which just failed to reach significance (i.e., a trend) on the presence of a natural mentor on youth outcomes. The strongest relations were found for the academic domain (r = .168); social confidence (r = .155); and physical health (r = .142). The risk-status (general or at-risk population) did not moderate the relation between natural mentorship and youth outcomes, indicating that the effect of the presence of a natural mentor does not differ for risk-status. For type of mentor, it was found that the percentage professional mentors significantly moderated the relation between presence of a natural mentor and youth outcomes. For ease of interpretation, the unstandardized regression coefficient in Table 4 was standardized (β = .221). The beta-coefficient indicates that having a professional mentor was moderately associated with more positive youth outcomes.
Considering study characteristics, the impact factor of the journal had a moderating trend-like effect on the relation between natural mentorship and youth outcomes. Again, for
26 ease of interpretation the unstandardized regression coefficient was standardized (β = .233). Stronger relations between the presence of a natural mentor and youth outcomes were found for studies published in journals with a higher impact factor. Publication year, study design, type of reporter, type of measure, uni- or multivariate effect, and the reliability of the assessment did not significantly influence the strength of the association between natural mentorship and youth outcomes. Also, no sample characteristics moderated the relation between the presence of a natural mentor and youth outcomes. Sex, the proportion of participants from ethnic minority groups in the sample, age, school type – and population did not moderate the relation between the presence of a natural mentor and youth outcomes.
Table 4
Moderators of the Relation between the Presence of a Natural Mentor and Youth Outcomes Moderator variable k #ES B0/
mean r t0 B1 t1 F(df1, df2) IV and DV characteristics Domain F(8, 157) = 1.961+ Academic (RC) 11 36 .168 4.905*** Social-emotional Social competence 7 15 .038 0.784 -.130 -2.278* Social confidence 9 17 .155 3.493*** -.013 -0.234 Vocational 8 16 .057 1.183 -.114 -1.971+ Physical health 4 14 .142 2.982** -.027 -0.483 Psychopathology Internalizing 15 28 .076 2.166* -.092 -1.981* Conduct problems 8 21 .095 2.450* -.073 -1.500 Overall psychopathology 3 3 .032 0.318 -.136 -1.274 Substance use 8 16 .018 0.407 -.151 -2.909** Study characteristics Publication year 22 163 .100 4.472*** -.002 -0.583 F(1, 161) = 0.339 Impact factor 22 163 .105 5.225*** .031 1.951+ F(1, 161) = 3.809+ Study design F(1,164) = 0.319 Cross-sectional (RC) 13 99 .087 3.017** Longitudinal 11 67 .111 3.444*** .024 .0565 Type of reporter F(4, 161) = 0.585 Self-report (RC) 21 149 .099 4.316***
Other report/ Teacher-report 3 6 .105 0.748 .006 0.045 Official registration 2 11 .155 1.930* .059 0.688
27 Single item (RC) 13 58 .116 3.794*** Multiple items 5 16 .051 1.049 -.065 -1.237 Scale 18 74 .097 3.557*** -.020 -0.560 Index 2 4 .067 0.764 -.050 -0.557 Reliability 16 68 .090 4.350*** -.048 -0.302 F(1, 66) = 0.091 Uni-/ multivariate F(1, 164) = 0.053 Univariate (RC) 17 133 .096 3.927*** Multivariate 7 33 .106 2.750** .010 0.043 Participant characteristics % ethnic minority 22 159 .103 4.453*** .000 0.042 F(1, 157) = 0.002 Ethnicity F(1, 164) = 0.223 Mixed sample (RC) 16 122 .105 4.047*** Minority sample 7 44 .082 2.035* -.023 -0.472 % male sample 22 159 .102 4.512*** -.001 -0.462 F(1, 157) = 0.2130 Sex F(1, 164) = 0.219 Mixed gender (RC) 21 150 .096 4.209*** Girls 3 16 .126 2.007* .030 0.468 Age 22 165 .097 4.424*** -.007 0.807 F(1, 163) = 0.651 Sample type F(1,164) = 0.021 Risk status (RC) 14 94 .101 3.514*** General population 9 72 .094 2.752** .007 0.114 Risk F(2, 163) = 0.936 Low (RC) 9 72 .094 2.933** Medium 7 48 .068 1.839* -.027 -.549 High 8 46 .135 3.645*** .042 .847 School type F(3, 141) = 0.526 Middle school (RC) 7 10 .079 0.905 High school 11 95 .098 3.061** .020 0.215 University 1 3 -.055 -0.396 -.133 -0.817 Mixed (school and no school) 5 37 .122 2.689** .045 .458
School population F(1, 143) = 0.691 Regular (RC) 19 130 .105 4.198*** Learning disabilities 1 15 .027 0.304 -.078 -0.831 Mentor/ relationship characteristics Mentor age 5 39 .112 4.136*** -.011 -1.500 F(1, 37) = 2.250 Percentage female 8 57 .088 2.731** .000 0.033 F(1, 55) = 0.001 Percentage kin 17 122 .098 3.930*** .002 0.200 F(1, 120) = 0.040 Percentage non-kin 14 102 .093 3.163** -.000 -0.360 F(1, 100) = 0.130 Percentage professional 14 101 .112 3.765** .002 2.055* F(1, 99) = 4.223* Length relationship 8 78 .102 2.411* .006 0.501 F(1, 76) = 0.251 Informational support 8 53 .129 2.441* .003 0.880 F(1, 51) = 0.775 Instrumental support 6 48 .128 1.974 -.001 -0.470 F(1, 46) = 0.221 Emotional support 8 55 .126 2.433* -.001 -0.391 F(1, 53) = 0.153 Ethnicity matched 5 21 .059 0.947 .009 1.492 F(1, 19) = 2.225 Gender matched 5 30 .049 2.074* -.007 -1.629 F(1, 28) = 2.652 Amount of contact F(1,60) = 1.964
28 Predominantly daily (RC) 3 10 .129 2.924**
Predominantly weekly 6 52 .060 2.228* -.070 -1.402
Feelings of closeness F(1, 33) = 1.008
Predominantly very close (RC)
6 31 .081 2.947**
Predominantly quite close 1 4 -.001 -0.009 -.082 -1.004
Note. IV and DV characteristics = independent variable (IV) and/ or dependent variable (DV); k = number of independent studies; #ES = number of effect sizes; B0/ mean r = intercept/ mean effect size (r); t0 =difference in mean r with zero; B1 =estimated regression coefficient; t1 = difference in mean r with reference category; F(df1, df2) = omnibus test; (RC) = reference category. + p ≤ .10. * p ≤ .05. ** p ≤ .01. *** p ≤ .001.
3.3. Overall relation between the quality of natural mentoring relationships and youth
outcomes
The meta-analysis on the relation between the quality of natural mentoring relationships and youth outcomes contains 8 independent studies (k), reporting on 56 effect sizes (#ES), and a total sample of N = 7363 participants. A small, significant relation was found between the quality of natural mentoring relationships and youth outcomes (r = .160; 95% CI = .052; .268; p = .004), indicating that the quality of natural mentoring relationships was significantly associated with more positive youth outcomes.
When checking for publication bias, Egger’s test was significant (t = 3.054, p = .003), indicating that there was funnel plot asymmetry (Egger et al., 1997). Next, a trim and fill procedure was conducted. This procedure yielded one missing effect size on the right side of the funnel plot, this is shown in Appendix 2. When taking this missing effect size into consideration, the overall effect became r = .174 instead of r = .160. These results suggest a marginal underestimation of the true effect.
The results of the likelihood-ratio tests showed that there was significant variance between effect sizes from the same study (i.e., level 2 variance) and that there was significant variance between studies (i.e., level 3 variance). Since the variances at level two and three
29 were significant, it was concluded that there was heterogeneity among the effect sizes that may be explained by characteristics of natural mentoring relationships, studies and samples. Therefore, moderator analyses were conducted.
3.4. Moderator analyses on the relation between the quality of natural mentoring
relationships and youth outcomes
Table 5 presents results from the moderator analyses on the relation between the quality of natural mentoring relationships and youth outcomes. The type of outcome domain (social-emotional; academic; vocational; psychopathology, substance use) significantly moderated the influence of the quality of natural mentoring relationships on youth outcomes. The largest effect was found for the vocational domain, although this was based on only one effect size (r = .314). An equal strong relation was found for the social-emotional domain (r = .280). A negative relation was found for substance use (r = -.233). The risk-status (general or at-risk population) did not moderate the relation between the quality of natural mentoring relationships and youth outcomes, indicating that the effect of the quality of natural mentoring relationships did not differ for risk-status. For type of mentor, it was found that the percentage professional mentors significantly moderated the relation between the quality of natural mentoring relationships and youth outcomes. For ease of interpretation this unstandardized regression coefficient was standardized (β = -.02). The beta indicated that having a professional mentor was negatively associated with youth outcomes.
Considering study characteristics, none of the moderators had a significant effect on the relation between the quality of natural mentoring relationships and youth outcomes. Also, no sample characteristics moderated the relation between the quality of natural mentoring relationships and youth outcomes. Sex, the proportion of participants from ethnic minority
30 groups in the sample, age, school type – and population did not moderate the relation between the quality of natural mentoring relationships and youth outcomes.
Table 5
Moderating Variables of Relation between the Quality of Natural Mentoring Relationships and Youth Outcomes
Moderator variable k #ES B0/ mean r t0 B1 t1 F(df1, df2) IV and DV characteristics Domain F(4, 51) = 18.482*** Psychopathology (RC) 3 19 .009 0.141 Substance use -.233 -4.184*** -.243 -2.938** Academic 4 11 .161 3.486** .151 2.202* Vocational 1 1 .314 2.692** .305 2.273* Social-emotional 5 25 .280 7.536*** .271 3.651*** Study characteristics Publication year 8 56 .145 2.452** -.006 -0.522 F(1, 54) = 0.273 Impact factor 8 56 .161 2.667* .002 0.024 F(1, 54) = 0.000 Study design F(1, 54) = 2.567 Cross-sectional (RC) 5 21 .213 3.255*** Longitudinal 3 35 .052 0.664 -.164 -1.602 Type of reporter F(1, 54) = 0.737 Self-report (RC) 8 53 .153 2.786** Official registration 2 3 .069 0.639 -.085 -0.859 Measurement F(1, 54) = 0.095 Continuous of availability (RC) 3 33 .126 1.348+ Quality of relationship 5 23 .163 2.195* .037 0.308 Type of measure F(2, 50) = 1.637 Single item (RC) 4 16 .080 1.297 Multiple items 2 2 .235 2.063* .160 1.279 Scale 8 35 .160 3.255*** .082 1.641 Reliability 8 35 .145 2.693** -.460 -1.238 F(1, 33) = 1.532 Uni/ multivariate F(1, 54) = 0.340 Univariate (RC) 7 46 .062 0.384 Multivariate 1 7 .162 2.612** .101 0.583 Participant characteristics % ethnic minority 8 56 .150 2.429** -.000 -0.023 F(1, 54) = 0.001 Ethnicity F(1, 54) = 0.463 Mixed sample (RC) 5 40 .114 1.448+ Minority sample 3 16 .190 2.262* .079 0.680 % male sample 7 54 .142 2.433** -.002 -0.652 F(1, 54) = 0.426 Sex F(2, 53) = 0.609 Mixed (RC) 6 49 .124 1.885* Boys only 1 5 .152 0.918 .028 0.157 Girls only 1 2 .337 1.863* .213 1.103
31 Age 8 56 .146 2.579** -.017 -0.576 F(1, 54) = 0.332 Sample type F(1, 59) = 0.084 General population (RC) 5 42 .136 1.869* F(1, 54) = 0.085 At risk population 3 14 .171 1.754* .036 0.292 Risk F(2, 53) = 0.620 Low (RC) 5 42 .136 1.889* Medium 2 12 .106 0.913 -.031 -0.228 High 1 2 .325 1.859* .200 1.023 School type F(2, 53) = 0.315 Middle school (RC) 2 16 .229 1.903* High school 5 35 .117 1.550+ -.114 -0.793 Mixed (school and no
school) 1 5 .152 0.882 -.080 -0.378 Mentor/ relationship characteristics Percentage female 2 9 .106 1.193 .011 1.077 F(1, 7) = 1.160 Percentage kin 5 20 .195 2.940** -.002 -0.547 F(1, 18) = 0.300 Percentage non-kin 3 11 .210 1.848* -.005 -0.547 F(1, 9) = 0.300 Percentage professional 4 28 .024 0.244 -.004 -1.840 F(1, 26) = 3.384* Length relationship 2 9 .106 1.193 -.017 -1.077 F(1, 7) = 1.160
Note. IV and DV Characteristics = independent variable (IV) and/or dependent variable (DV); k = number of independent studies; #ES = number of effect sizes; B0/ mean r = intercept/ mean effect size (r); t0 =difference in mean r with zero; B1 =estimated regression coefficient; t1 = difference in mean r with reference category; F(df1, df2) = omnibus test; (RC) = reference category. + p ≤ .10. * p ≤ .05. ** p ≤ .01. *** p ≤ .001. Discussion
This is the first meta-analytic study assessing the relation between natural mentoring relationships and youth outcomes. The review included 30 studies on the topic dating from 1992 to present. Two separate meta-analyses were conducted on the relation between the presence of a natural mentor and youth outcomes, and natural mentoring relationship quality and youth outcomes. Small significant associations were found, with r = .097 for the presence of a natural mentor, and r = .160 for the quality of natural mentoring relationships (Cohen, 1992). The results highlight the importance of relationship quality, indicating that the presence of a natural mentor was related to positive youth outcomes and that a higher quality
32 of the natural mentoring relationship was somewhat stronger correlated with positive youth outcomes. In the meta-analysis on the relation between the presence of a natural mentor, the strongest relation was found for the academic domain (r = .168). In the meta-analysis on the relation between the quality of natural mentoring relationships and youth outcomes, the strongest relations were found for the vocational (r = .314) and social-emotional domain (r = .280).
The finding that the presence of a natural mentor is related to positive youth outcomes is in line with the conclusions drawn in the systematic review of natural mentoring in foster care by Thompson et al. (2016). Since the current meta-analysis included nationally representative samples as well as specific samples (e.g., risk, minority youth) the findings are applicable to adolescents in general compared to the review by Thompson et al. (2016), which exclusively focused on youth in foster care. Moreover, the (narrative) review by Thompson was qualitative instead of quantitative, and therefore did not compute effect sizes.
When looking at the outcome domains, the relatively strong effects for the academic domain are consistent with previous meta-analyses and studies on formal mentoring relationships (DuBois et al., 2011; Karcher, Davis, & Powell, 2002; Raposa et al., in press), and may reflect the particular salience of caring teachers, guidance counselors and other natural mentors in educational settings. Furthermore, the positive findings for the quality of natural mentoring relationships were as expected. These findings are consistent with a meta-analysis on mentoring relationships in general (i.e., formal and natural relationships), which showed that when being in a high-quality relationship the amount of support increases (Eby et al., 2013). Developing high quality relationships requires spending time to getting to know each other. The more frequently the mentor and the youth interact, and the more satisfying the relationship is, the greater the opportunity for the mentor to provide the youth with support
33 (Huston & Burgess, 1979). In turn, this reinforces the expectation a youth has for the provision of support to fulfill his needs (Ragins & Verbos, 2007).
In the first meta-analysis, the strongest association was found for professional mentors. This relatively strong effect is consistent with a meta-analysis on teacher-student relationships (Roorda, Koomen, Spilt, & Oort, 2014). Further, this finding is in line with a single study on natural mentoring characteristics (DuBois & Silverthorn, 2005b), showing that youths can benefit more from relationships with natural mentors outside the familial system through the working mechanism of building social capital (Darling, Hamilton, & Niego, 1994). However, the second meta-analysis showed a negative association between quality of the relationship with a professional mentor and positive youth outcomes. However, a thorough examination revealed that the negative association was unduly affected by a study examining the effects of professional mentors on school attachment and risk behaviors (Black, Grenard, Sussman, & Rohrbach, 2010). This study showed that professional mentors had a protective indirect influence on risk behavior through a positive association with school attachment. The negative association may be explained by the fact that teachers as natural mentors and youth only interact in the school-context, making it difficult for a professional mentor to have a direct effect outside the school context (Roorda, Koomen, Spilt, & Oort, 2011).
It was expected that risk-status would moderate the relation between natural mentoring relationships and outcomes, because it may be difficult for youth who face numerous challenges (e.g., academic problems, parental conflict, low SES) to overcome these challenges with the presence of a natural mentor alone. However, risk-status did not prove to be a significant moderator. The absence of a significant moderating effect may indicate that natural mentors are generally beneficial for all youth regardless of risk-status. On the one hand, natural mentors may serve as complementary resources and promote PYD when youth
34 have good familial relationships. On the other hand, natural mentors may serve as complementary resources for youth from backgrounds of risk or for whom parents may not be fully engaged in their lives (Britner, Balcazar, Blechman, Blinn-Pike, & Larose, 2006; DuBois, Holloway, Valentine, & Cooper, 2002; Erickson et al., 2009). Although youth with a history of unavailable or inconsistent care may be less likely to turn to others in times of stress (Belsky & Cassidy, 1994), natural mentors who are trustful and consistent in their relationship with the youth may help them feel worthy and become more open for emotional support to cope with stressful events or chronic adversity (Rhodes et a., 2006; Rutter, 1987).
The findings of the two meta-analyses in this study indicate that the quality of natural mentoring relationships is crucial for understanding how natural mentoring relationships may have a positive effect on youth outcomes. Besides the presence of a natural mentor, for most outcome domains the effects are more positive when quality of the natural mentoring relationship is higher (e.g., social-emotional outcomes). Interestingly, meta-analyses on the therapeutic alliance in youth care (Karver, Handelsman, Fields, & Bickman, 2006; McLeod, 2011; Roest, Welmers – van de Poll, van der Helm, Hoeve, & Stams, in press; Shirk & Karver, 2003, 2011; Shirk, Karver, & Brown, 2011), with overall effect sizes ranging between r = .14 and r = .22 the present meta-analysis shows a similar effect on youth outcomes when
having a high-quality natural mentoring relationship. Taking advantage of these natural mentoring relationships in youth (health) care could possibly increase the effects of treatment and youth care interventions targeting troubled youth (in curative interventions) or youth at risk for negative developmental outcomes (in preventive interventions) by securing continuity in cure and care, fostering treatment motivation, and taking full advantage of the natural social network of the child (Van Dam et al., in press; Spencer et al., 2016). However, a serious caveat of taking advantage of the natural social network in youth care, in particular the informal mentor, is the natural paradox, that is: how can we profit from the natural mentoring
35 relationship without professionalizing it? In other words: how to strike the balance between using and supporting informal mentoring relationships on the one hand, and appreciating the natural working mechanisms of the informal mentoring relationship on the other hand?
The Youth Initiated Mentoring approach is a new approach in which youth recruit a caring adult from within their community as (natural) mentor (Schwartz et al., 2013). The YIM-approach helps and guides youth in the process of selecting and recruiting a natural mentor, making it even possible for high-risk youths to select a mentor (Van Dam et al., in press; Spencer et al., 2016). In depth case studies illustrate the natural paradox: some YIM's say they want to be involved, but do not want to be approached as a natural mentor. They want to stay neighbor, uncle or friend of the family, whereas others say it helps them to shift in position: 'as mentor, I feel legitimated to be more direct in my advice then when I'm grandfather' (Van Dam, Bakhuizen, Van Gelder, Stams, & Wissink, 2017). An advantage of this approach is that it makes use of already existing relationships, which is in line with research indicating that strong emotional connections between youth and a mentor are found to be important relationship features related to better youth outcomes (DuBois & Neville, 1997). Also, YIM-relationships appear to be long-lasting, and may be an addition to formal mentoring relationships, where long waitlists exist due to difficulties with the recruitment of volunteer mentors (Rhodes, 2002; Schwartz et al., 2013; Spencer et al., 2016). Sensitivity and systemic knowledge is needed to find a balance in this paradoxical situation of using natural mentors in youth care.
Results of the meta-analytic study may have implications for youth policy in Western countries, such as Northern-America and Europe, since the results emphasize the importance of creating opportunities for natural mentoring relationships to develop. For example, this can be done by strengthening teacher-student relationships in schools, which have been shown to have a substantial and positive impact on students’ academic achievement (Roorda et al.,
36 2011), behavior problems (Lei, Cui, & Chiu, 2016; Van der Werff et al., in press), and social-emotional development (Roelofs et al., in press), providing sufficient opportunities for youth participation in extracurricular informal learning activities, and investment in high quality youth work (Clarijs, 2008; European Commission, 2015).
The results of this meta-analytic study must be viewed within the context of the limitations associated with the empirical studies on which this meta-analysis is based. First, a non-ambiguous and well-validated definition of what constitutes natural mentoring relationships is absent in the field of natural mentoring, which makes it difficult to include all available quantitative studies (Thompson et al., 2016). For example, in some studies natural mentors had to be at least twenty years; five years older than the youth; or known to the youth for at least two years (Ahrens, DuBois, Lozano, & Richardson, 2010; Hurd & Zimmerman, 2010b; Rhodes, Ebert, & Fischer, 1992). In this way, peers could not be included as natural mentors. However, peers may be considered as one of the three primary socialization sources for youth during adolescence (Oetting & Donnermeyer, 1998). Peers form close relationships with youths, and may provide the youth with appropriate support, since they face the same developmental tasks (Karcher et al., 2005; Whitney et al., 2011). In the study by Whitney et al. (2011), the effects of peer mentoring on self-esteem were even larger when compared to adult mentors. Second, risk-status could not unambiguously be defined, since some studies were very small in focus (e.g., pregnant and parenting teenagers, homeless youth, children of alcoholics). This impairs the generalizability of the findings, and the possibility to further examine the effects of risk-status in terms of environmental or individual risk. Consequently, the results regarding the relation between natural mentoring and risk-status deserve careful interpretation. Third, the hypothesis on type of mentor could not be fully tested, since there was only one study in the field of natural mentoring testing the effects of various types of mentors (e.g., kin-mentor versus non-kin mentor) on youth outcomes (DuBois & Silverthorn,
37 2005b). Also, the studies testing a specific type of mentor compared to no mentor were limited in number (N = 4).
There are some limitations of the current study that need to be mentioned. First, for the meta-analysis on the presence of a natural mentor, 46 effect sizes were estimated to be missing on the right side of the funnel plot when checking for publication bias. For the second meta-analysis one positive result was missing (see Appendix 2). The missing effect sizes resulted in an underestimation of the true effect size, and therefore should be interpreted as selection bias given that missing studies that result in a smaller overall effect size would indicate publication bias. Selection bias may be due to an overrepresentation of certain samples or groups (e.g., data from the Adolescent Health Study). Second, there are limitations with respect to the generalizability of the study findings. The total sample consisted of 24 studies for the first meta-analysis, and only eight studies for the second meta-analysis, which should lead to careful interpretation of the overall findings. Notably, some moderator analyses were based on a small number of effect sizes (i.e., less than three effect sizes or studies). This implies low statistical power to detect the contributions of moderator variables and restricted generalizability. Last, all studies were conducted in the USA. This limits the possibility to draw general conclusions for other countries or continents.
An implication for future research is conducting research on more important life domains of adolescents. This should not be limited to psychological and behavioral outcomes, but expands to the relationships youth have with important persons in their life. For example, relationships with friends are an important part of the development of youth, and have been shown to be associated with the social, emotional and cognitive development of adolescents (Newcomb & Bagwell, 1995; Vaquera & Kao, 2008). It is important to determine if a natural mentor influences the relationships youth have with peers and significant others. Leisure time and extracurricular youth activities are also an understudied topic when focusing on the
38 influence of natural mentors on youth outcomes. Research on a broader range of outcome domains may provide knowledge on the type of outcomes where natural mentoring is most influential. This could inform researchers and policy makers about the type of goals that natural mentoring might aim for with the greatest chance of success.
Furthermore, meaningful relationships are not restricted to natural mentoring relationships alone. Research should include the social support framework too, since natural mentoring can be considered as a specific form of social support (Zimmerman et al., 2002). In this way, more potential sources of social support (e.g., parents, family members, peers, community members, teacher) may be included when conducting systematic research, resulting in a broader range of available studies.
Last, mentor characteristics (e.g., psychological well-being, deviant behavior, having a job, school completion) should be studied more systematically to determine which characteristics of the natural mentor result in a successful (i.e., supportive or protective) relationship or not. Notably, when natural mentors are engaged in problem behavior (i.e., substance use, delinquency), youths are more likely to be negatively affected by such deviant behaviors, which may be ascribed to negative role-modelling (Chen, Greenberger, Farruggia, Bush, & Dong, 2003; Sterrett, Jones, McKee, & Kincaid, 2011).
Perhaps the most important lesson to be learned from this study is that natural mentoring matters. However, the presence of a natural mentor may not be enough. When aiming at achieving significant changes in the lives of youth, relationship characteristics and quality were found to be crucial. By intervening on the quality of the relationship, the natural mentoring relationship may become even more effective. However, it should be kept in mind not to change the nature of the already existing relationship, turning it into a formal relationship. The findings provide hope about the capacity of natural mentors to improve and
39 even transform young lives. These initial findings challenge us to further understand the complexities of natural mentoring.
40
Appendix 1
281 titles screened and abstracts read in ERIC, PsychINFO, PubMed, Wiley Online Library, Google Scholar, and ResearchGate.
Reasons for exclusion:
• Adult mentoring (mean age >24 year): 59 • Mentoring program: 56 • School mentors: 49 • Career mentors: 26 • E-mentoring: 5 • No youth outcomes: 49 • Double articles: 3 • Parent mentoring: 1 Total excluded: 248
Ancestry method on 33 articles
Included, based on following subjects:
• Youth mentoring and youth outcomes: 19 • VIP and youth outcomes: 4
• Important nonparental adult and youth outcomes: 8 • Natural mentoring and youth outcomes: 7
Included, by contacting authors: • Rhodes: 1
Total included: 39
72 full texts thoroughly examined and methods briefly screened Reasons for exclusion:
• Systematic review: 2 • Qualitative studies: 20 • Adult mentoring: 2
• Mentorship characteristics: 7
• Mentoring program with informal mentors: 3 • Meta-analysis on formal mentoring: 2
• Mentor perspective studies: 1 • Formal mentoring: 1
• Not suitable for analysis: 4 Total excluded: 42
