ArticlePDF Available

Effectiveness of Mentoring Programs for Youth: A Meta-Analytic Review

Authors:

Abstract

We used meta-analysis to review 55 evaluations of the effects of mentoring programs on youth. Overall, findings provide evidence of only a modest or small benefit of program participation for the average youth. Program effects are enhanced significantly, however, when greater numbers of both theory-based and empirically based "best practices" are utilized and when strong relationships are formed between mentors and youth. Youth from backgrounds of environmental risk and disadvantage appear most likely to benefit from participation in mentoring programs. Outcomes for youth at-risk due to personal vulnerabilities have varied substantially in relation to program characteristics, with a noteworthy potential evident for poorly implemented programs to actually have an adverse effect on such youth. Recommendations include greater adherence to guidelines for the design and implementation of effective mentoring programs as well as more in-depth assessment of relationship and contextual factors in the evaluation of programs.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
American Journal of Community Psychology, Vol. 30, No. 2, April 2002 (
C°
2002)
Effectiveness of Mentoring Programs for Youth:
A Meta-Analytic Review
David L. DuBois,
1
Bruce E. Holloway, Jeffrey C. Valentine,
and Harris Cooper
University of Missouri at Columbia
We used meta-analysis to review 55 evaluations of the effects of mentoring
programs on youth. Overall, findings provide evidence of only a modest or
small benefit of program participation for the average youth. Program effects
are enhanced significantly, however, when greater numbers of both theory-
based and empirically based “best practices” are utilized and when strong re-
lationships are formed between mentors and youth. Youth from backgrounds
of environmental risk and disadvantage appear most likely to benefit from par-
ticipation in mentoring programs. Outcomes for youth at-risk due to personal
vulnerabilities have varied substantially in relation to program characteris-
tics, with a noteworthy potential evident for poorly implemented programs
to actually have an adverse effect on such youth. Recommendations include
greater adherence to guidelines for the design and implementation of effective
mentoring programs as well as more in-depth assessment of relationship and
contextual factors in the evaluation of programs.
KEY WORDS: youth mentoring; program evaluation; primary prevention; children;
adolescents.
INTRODUCTION
During the past decade mentoring programs for youth have become
increasingly popular and widespread. Big Brothers/Big Sisters of America
(BB/BSA), the most prominent of these programs, now includes over 500
agencies nationwide. The National Mentoring Partnership and numerous
1
To whom correspondence should be addressed at 210 McAlester Hall, Department of
Psychology, Universityof Missouri atColumbia, Columbia, Missouri65211; e-mail: DuBoisD@
missouri.edu.
157
0091-0562/02/0400-0157/0
C
°
2002 Plenum Publishing Corporation
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
158 DuBois, Holloway, Valentine, and Cooper
other organizations also have contributed to significant growth in mentor-
ing initiatives at local, state, and national levels (Johnson & Sullivan, 1995).
Currently, the National Mentoring Database lists more than 1,700 organiza-
tions that support mentoring activities (Save the Children, 1999).
Interest in mentoring programs has been fueled in significant part by the
importance that positive relationships with extrafamilial adults have been
indicated to have in promoting resiliency among youth from at-risk back-
grounds (Rhodes, 1994). It should not be assumed, however, that the essen-
tial features of these types of naturally occurring relationships can reliably
be reproduced by programs that seek to provide youth with adult men-
tors through necessarily more artificial mechanisms (Hamilton & Hamilton,
1992). Studies evaluating the benefits of mentoring programs for youth
have begun to appear only recently in the literature. Prior reviews (Darling,
Hamilton, & Niego, 1994; Flaxman, Ascher, & Harrington, 1988; Johnson
& Sullivan, 1995; Rhodes, 1994), therefore, have been limited by a lack of
available data upon which to base conclusions. Furthermore, because of
multidisciplinary and applied interest in mentoring, reports have appeared
in diverse literatures and a significant proportion have been published pri-
vately by foundations and other organizations.
The present research utilizes meta-analysis to review and synthesize the
existing empirical literature on youth mentoring programs (Cooper, 1998;
Durlak & Lipsey, 1991). Meta-analysis offers several advantages over the
narrative approach that has been employed in prior reviews. These include
(a) explicit operationalization of literature search procedures to help re-
duce omissions or bias in the investigations that are identified for review;
(b) an objective and quantifiable basis for assessing the overall magnitude
of program effects on youth; and (c) the ability to test for significant differ-
ences in findings across investigations along any dimension of interest, thus
facilitating identification of factors that may have important implications for
program effectiveness. This latter concern seems particularly germane to the
study of youth mentoring programs given the considerable diversity that has
characterized intervention efforts in this area (Rhodes, 1994). Factors merit-
ing consideration as sources of influence on the results of mentoring program
evaluations include (a) features of program design and implementation;
(b) characteristics of participating youth; (c) qualities of the mentor–mentee
relationships that are formed; and (d) issues relating to the assessment of
youth outcomes.
Program Design and Implementation
From a program design standpoint, many programs (e.g., BB/BSA)
have focused solely on providing mentoring relationships to youth. In other
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 159
instances, mentoring has been implemented as one of several distinct compo-
nents of a multifaceted intervention program. Enhanced benefits generally
have been expected to result when mentoring is linked to other supportive
services (Flaxman et al., 1988). Nevertheless, there also may be certain ad-
vantages to program specialization in mentoring. With regard to this latter
possibility, BB/BSA has been widely discussed as a model of “best practices”
for youth mentoring (e.g., Tierney, Grossman, & Resch, 1995). The effective-
ness of this program relative to non-BB/BSA programs is thus of particular
interest.
Mentoring programs also have differed in their basic goals and phi-
losophy. Thus, whereas some programs have pursued the general goal of
promoting positive youth development, others have adopted more focused
or instrumental goals relating to areas such as education or employment
(Saito & Blyth, 1992). The relative merits of these contrasting program
orientations has attracted a considerable amount of discussion in the lit-
erature (Darling et al., 1994; Freedman, 1992; Hamilton & Hamilton, 1992;
Tierney et al., 1995), with arguments offered in favor of each type of
approach.
Further considerations pertain to the procedures used for recruiting
prospective mentors and the levels of training and supervision that are
provided to mentors once selected (Rhodes, 1994). Background checks
and other screening procedures (e.g., interviews) have been included
consistently in recommended guidelines for the selection of mentors in
programs (Freedman, 1992; National Mentoring Working Group, 1991;
Saito & Blyth, 1992). Some programs also have specifically sought out
individuals whose backgrounds (e.g., teacher) may make them especially
well-suited to forming effective mentoring relationships with youth. There
has been less consensus regarding needs for training and ongoing
supervision of mentors (Rhodes, 1994) and accordingly programs have
varied considerably in these areas. Nevertheless, there is general agree-
ment that some type of orientation should be provided and that mentors
should have ongoing support available to them (Freedman, 1992; Hamilton
& Hamilton, 1992; National Mentoring Working Group, 1991). Additional
recommendations include matching of youth with mentors on the basis
of criteria such as gender, race/ethnicity, or mutual interests; communica-
tion of guidelines and expectations regarding frequency of mentor–mentee
contact and duration of relationships; monitoring fidelity of implementa-
tion through mentor logs and other procedures; incorporation of struc-
tured opportunities for mentor–mentee interaction; and provisions for
the support and involvement of parents (Freedman, 1992; Hamilton &
Hamilton, 1992; National Mentoring Working Group, 1991; Saito & Blyth,
1992).
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
160 DuBois, Holloway, Valentine, and Cooper
Characteristics of Youth
The significance attached to mentoring relationships as a protective in-
fluence suggests that programs may provide greater benefits to youth who
can be considered “at-risk” by virtue of individual and/or environmental cir-
cumstances (Rhodes, 1994). Accordingly, these youth have been the focus
of a large proportion of mentoring programs (Freedman, 1992) and cur-
rently constitute the majority of all those receiving mentoring (McLearn,
Colasanto, & Schoen, 1998). Other specific subgroups that have been tar-
geted by programs include youth from single-parent homes (e.g., BB/BSA)
and those belonging to racial or ethnic minority groups (e.g., Royse, 1998).
Programs also have been directed toward youth of varying ages and devel-
opmental levels. Possible sources of influence on outcomes in this regard in-
clude the optimal timing of mentoring as a preventive intervention (Institute
of Medicine, 1994) as well as practical issues pertaining to implementation
(e.g., receptivity of youth to mentoring at differing stages of development).
Mentor–Mentee Relationships
In order to yield desired outcomes, it may be necessary for programs to
establish mentoring relationships between youth and adults that involve pat-
terns of regular contact over a significant period of time (DuBois & Neville,
1997; Freedman, 1992; Slicker & Palmer, 1993). Realization of this aim can be
limited, however, in actual practice by difficulties encountered in the recruit-
ment of needed mentors, inadequate levels of mentor–mentee involvement,
and premature termination of relationships prior to fulfillment of program
expectations (Freedman, 1992; Hamilton & Hamilton, 1992). The extent to
which mentoring relationships with consistent and sustained patterns of in-
teraction are actually formed in programs therefore represents a potentially
important source of variation in outcomes. A related, methodological con-
sideration is whether youth with relationships that fail to meet criteria for
minimum levels of contact or longevity are excluded from analyses of pro-
gram effectiveness. When this is done the result may be an unduly positive
assessment of the benefits that can be realistically expected for all youth
referred to a given mentoring program (Grossman & Tierney, 1998).
Assessment of Outcomes
Mentoring programs have been conceptualized as potentially affect-
ing youth in a wide variety of areas, including emotional and behavioral
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 161
functioning, academic achievement, and employment or career develop-
ment. One important concern therefore is whether benefits of mentoring
are evident across this diverse range of proposed outcomes. Further consid-
erations include the type of data source or informant utilized as well as the
timing of outcomes assessment relative to the active period of program op-
eration. To the extent that effects on youth are evident both across multiple
sources of data and at follow-up assessments, this would provide particularly
strong support for the effectiveness of mentoring programs.
This Study
The specific aims of this study are two-fold: (a) to objectively assess
the overall effects of mentoring programs on youth and (b) to investigate
possible variation in program impact in association with factors relating to
each of the aforementioned areas (i.e., program design and implementa-
tion, youth characteristics, mentor–mentee relationships, and assessment of
outcomes). The primary goal of the latter analyses is to help identify promis-
ing directions for enhancing program effectiveness. Both theory-based and
empirically based indices of best practices for mentoring interventions are
developed for use in this portion of the research. These indices are utilized in
an effort to identify specific constellations of program characteristics asso-
ciated with enhanced effectiveness. Because of the importance attributed
to relationship factors as moderators of program outcomes, supplemen-
tary analyses also are conducted of comparisons that have been made in
several studies within the intervention group on the basis of relevant fea-
tures or characteristics of the relationships formed between mentors and
youth.
METHOD
Literature Search Procedures
Three primary methods were used to locate evaluations of youth men-
toring programs. First, computer searches of PsychINFO, ERIC, Medline,
and Dissertation Abstracts reference databases were run using both subject
terms (e.g., mentor) and textwords (e.g., Big Brother) to identify relevant
articles. The time frame for each search was from 1970, when research on the
outcomes of mentoring programs began to appear, through 1998. Second, in
an effort to further identify possible unpublished evaluation studies, a search
of the Internet was conducted using several search engines (e.g., Yahoo!).
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
162 DuBois, Holloway, Valentine, and Cooper
Finally, the reference sections of reports that met criteria for inclusion in the
meta-analysis were examined to determine if reference was made to other
potentially relevant reports.
Criteria for Including Studies
To be included in the present review, studies needed to meet several
criteria. First, the program evaluated needed to involve mentoring as the
practice has been defined commonly in the literature (Nettles, 1991; Rhodes,
1994). To maintain consistency with the prevailing view of mentoring as en-
tailing a one-on-one relationship, programs in which mentoring appeared to
have occurred primarily on a small group basis were not included. Similarly,
because mentoring generally has been regarded as referring to a relation-
ship between an older, more experienced mentor and a younger protegee
(Rhodes, 1994), peer tutoring or mentoring programs were excluded from
the present review, although those in which older youth (e.g., teenagers)
served in a mentor capacity for younger children were eligible for inclusion.
Also excluded were those programs in which the adults involved in form-
ing relationships with youth were mental health professionals (e.g., social
workers).
2
Second, the study had to examine empirically the effects of par-
ticipation in a mentoring program, either by preprogram versus postprogram
comparison on the same group of youth or a comparison between one group
of youth receiving mentoring and another group not receiving mentoring
drawn from the same population. The decision to include evaluations report-
ing either of the two types of comparisons was based on a desire to increase
the number of studies available for review and hence enhance power in tests
of both overall effects of mentoring programs and possible moderators of
their effectiveness. Finally, the sample used in the evaluation of the program
needed to include youth with a mean age of less than 19. A decision also
was made to exclude from the review evaluations of two well-established
prevention programs that have included mentoring-related components in
their design: the Adolescent Diversion Project (Davidson & Redner, 1988)
and the Primary Mental Health Project (Cowen et al., 1996). Neither of these
programs focuses specifically on mentoring and both already have received
extensive consideration in prior literature reviews.
2
The criterion relating to use of mental health professionals as mentors resulted in the well-
known Cambridge–Somerville study being excluded from consideration because social work-
ers served as the companions to youth in this intervention (see McCord, 1992). The findings of
this intervention frequently are cited as an example of the potential for preventive interven-
tions to have unanticipated negative effects on participants; as will become apparent, however,
a similar lesson can be drawn from the results of those evaluation studies of mentoring pro-
grams that did meet criteria for inclusion in the present review.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 163
Search Outcome
Search procedures identified 59 separate research reports that met cri-
teria for inclusion and for which information was available to allow for the
computation of one or more effect sizes. One report (Tierney et al., 1995)
was simply an earlier, unpublished version of a large scale multisite evalu-
ation of BB/BSA programs that was subsequently published (Grossman &
Tierney, 1998). Two articles presented immediate posttest and then 2-year
follow-up findings for the same sample of youth participating in the Buddy
System mentoring program (Fo & O’Donnell, 1975; O’Donnell, Lydgate, &
Fo, 1979) and thus were considered for the purposes of the present review to
constitute only one independent study. Finally, two additional articles (Fo &
O’Donnell, 1974; Rhodes, Haight, & Briggs, 1999) presented findings on
subsamples of youth from the preceding, larger evaluations of BB/BSA and
the Buddy System program. These reports were excluded from the present
review to avoid overlap in samples across reports. On the basis of the fore-
going decisions, a total of 55 independent studies or reports were retained
for further analysis.
Effect Size Calculations
Effect sizes were computed as d-indexes, or standardized mean dif-
ferences (Cohen, 1988). The d-index expresses the difference between two
group means in terms of their common standard deviation. In the present
context, d-indexes were calculated both for comparisons of preprogram ver-
sus postprogram means for a given group of youth participating in a mentor-
ing program as well as for comparisons between one group of youth receiving
mentoring and another group not receiving mentoring. If information rel-
evant to both types of comparisons was available, separate d-indexes were
computed for each form of comparison. Whenever possible, d-indexes were
calculated from means and standard deviations provided by the report writ-
ers. When means and standard deviations were not provided but the values
of corresponding statistical tests of mean differences were given, formulas
provided by Rosenthal (1994) were used to estimate d-indexes. When nei-
ther type of information was reported, efforts were made to obtain relevant
data from the first author of the report. For a few reports (n = 7), the only in-
formation available to compute d-indexes was test statistics associated with
analyses (e.g., analysis of covariance) that controlled statistically for other
variables, such as pretest scores on the outcome measure. These reports
included several evaluations that were based on relatively large samples
of youth and that involved random assignment to intervention and control
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
164 DuBois, Holloway, Valentine, and Cooper
conditions. Because of these methodological strengths, a decision was made
to include them in the review by estimating d-indexes on the basis of the test
statistics available. Overall, across the 55 independent studies that met crite-
ria for review a total of 575 separate estimates of effect size (i.e., d-indexes)
were calculated.
Each effect size was weighted by the inverse of its variance to provide
more efficient estimation of true population effects (Hedges & Olkin, 1985).
This procedure gives greater weight to samples based on larger samples and
is the generally preferred alternative (Cooper, 1998). Variance estimates for
one-group preprogram versus postprogram effect sizes were based on the
formula v = [1 + (d
2
/2)]/n (Cooper, Charlton, Valentine, & Muhlenbruck,
2000). Variance estimates for two-group comparisons were calculated using
formulas given in Cooper (1998). For purposes of comparison, findings for
both unweighted and weighted effect sizes are reported in analyses of over-
all program effects. Only weighted d-indexes are analyzed and reported in
analyses of moderator variables. All effect sizes were coded so that positive
values indicated differences in directions consistent with a favorable effect
of the mentoring program.
Coding of Studies
Each report was coded on multiple characteristics. The characteristics
could be divided into six major categories: (a) report information (year of re-
port, published/unpublished); (b) evaluation methodology (type of research
design, use of statistical control variables, internal vs. external evaluation,
sample size, exclusion of nonactive relationships from analyses); (c) pro-
gram features (mentoring alone vs. mentoring as part of multicomponent
intervention, BB/BSA vs. non-BB/BSA, program goal, geographic location,
setting in which mentoring activities occurred, compensation of mentors,
monitoring of implementation, characteristics of mentors recruited, pro-
cedures for screening prospective mentors, criteria for matching mentors
and youth, mentor training, supervision, and support, expectations for fre-
quency of contact and length of mentoring relationships, structured activities
for mentors and youth, inclusion of parental support or involvement com-
ponent); (d) characteristics of participating youth (gender, race/ethnicity,
developmental level, single-parent household, socioeconomic background,
at-risk status); (e) mentor–mentee relationships (actual frequency of con-
tact, average length); and (f) assessment of outcomes (type of outcome, data
source, timing of assessment).
3
3
A copy of the coding sheet is available from the first author.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 165
The theory-based index of best practices referred to previously was de-
rived on the basis of the presence of the following 11 program features:
monitoring of program implementation, screening of prospective mentors,
matching of mentors and youth on the basis of one or more relevant cri-
teria, both prematch and ongoing training, supervision, support group for
mentors, structured activities for mentors and youth, parent support or in-
volvement component, and expectations for both frequency of contact and
length of relationships. Each of these program features has been included
in previous recommendations for establishing effective mentoring programs
(Freedman, 1992; Hamilton & Hamilton, 1992; National Mentoring Work-
ing Group, 1991; Saito & Blyth, 1992). The number of practices reported
for any given program served as its score on this index. The empirically
based index of best practices was derived in a similar manner, but was based
on those program features that reached or approached (p <.10) statistical
significance as moderators of effect size in the present investigation. The
program features eligible to contribute to this index included those compris-
ing the theory-based index as well as all other aspects of program design and
implementation that were examined as potential moderators (e.g., compen-
sation of mentors). It should be noted that the focus on significant individual
moderators of effect size in constructing the empirically based index did not
necessitate that a trend toward enhanced outcomes would be evident in
association with a greater overall number of the features involved being
characteristic of particular programs. For such a relationship to be found,
the different program features would need to exhibit, to a substantive ex-
tent, a nonoverlapping and hence incremental pattern of association with
estimates of effect size.
Unit of Analysis
For the present investigation, the independent sample was the primary
unit of analysis. Because effect size information was reported for the overall
sample in most reports, each report or study generally contributed one inde-
pendent sample to the analysis. If a study only reported findings separately
for different, nonoverlapping subgroups, however, such as boys and girls, it
contributed more than one sample to the analysis. Overall, the 55 reports
yielded a total of 60 independent samples for analysis.
Within this general framework, a shifting unit of analysis approach was
used for determining what constituted an independent estimate of effect
(Cooper, 1998). In this procedure, each effect size is first coded as if it were
an independent estimate of the intervention’s impact. For example, if data
for a single report permitted comparison of preprogram and postprogram
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
166 DuBois, Holloway, Valentine, and Cooper
scores for both problem behavior and school performance, two separate
d-indexes would be calculated. When estimating the overall effect of men-
toring, these two d-indexes would be averaged prior to entry into the analysis,
so that the sample contributed only one effect size. However, in an analy-
sis that examined the effect of mentoring on problem behavior and school
performance separately, the sample would contribute one effect size to esti-
mates of mean effect size for each of the two relevant categories of outcome
measure.
Fixed and Random Effects
A final consideration prior to conducting analyses was the need to de-
cide whether to conceptualize the effect of youth mentoring programs as
fixed or random (Hedges & Vevea, 1998). In a fixed-effect analysis, each
effect size’s variance is assumed to reflect only sampling error (i.e., error
solely due to participant differences). This source of error can be taken into
account through procedures described previously for weighting effect sizes
by sample size. When a random-effect analysis is carried out, a study-level
variance component is assumed to be present as an additional source of
random influence. Hedges and Vevea (1998) state that fixed-effect models
of error are most appropriate when the goal of the research is “to make
inferences only about the effect size parameters in the set of studies that
are observed (or a set of studies identical to the observed studies except for
uncertainty associated with the sampling of subjects)” (p. 3). In general, a
random-effect analysis is more conservative because of the consideration of
study-level variance as an additional component of error (Wang & Bushman,
1999).
The appropriateness of a random effects model in the present context is
suggested both by (a) the large variation that is evident in the implementation
and design characteristics of youth mentoring programs (and hence the po-
tential for these factors to constitute significant sources of random error even
after taking into account variance associated with specified moderating vari-
ables) and (b) interest in drawing inferences about all youth mentoring pro-
grams, not just those included in the present review. Alternatively, a fixed ef-
fects could be argued to be appropriate to the extent that the effectiveness of
those programs that have been subjected to evaluation is of particular inter-
est. Relevant considerations in this regard include the relatively widespread
dissemination that some of the most frequently evaluated programs have
received (e.g., Big Brothers/Big Sisters) as well as the possibility that pro-
grams undergoing formal evaluation may tend to be most representative
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 167
of “best practices” in youth mentoring (e.g., more innovative designs, greater
monitoring and fidelity in program implementation, etc.). A further statis-
tical consideration is that in the search for moderators, fixed-effect models
may seriously underestimate and random-effects models seriously overesti-
mate error variance when their assumptions are violated (Overton, 1998).
In view of the foregoing competing sets of concerns, a decision was made
to apply both models in all primary study analyses (Cooper et al., 2000).
Specifically, all analyses were conducted twice, once employing fixed-effect
assumptions and once using random-effect assumptions. This allowed simi-
larities and differences in results across the two types of analyses then to be
incorporated into the interpretation and discussion of findings. Both fixed-
effect and random-effect analyses were carried out using SAS Software in
accordance with procedures described by Wang and Bushman (1999).
RESULTS
Preliminary Analyses
Inspection of the distribution of the 575 unweighted d-indexes using a
stem-and-leaf plot revealed 11 positive d-index values that were more than
three interquartile ranges beyond the 75th percentile and thus qualified as
statistical outliers according to Tukey’s definition (Tukey, 1977). In addition,
two effect sizes met a corresponding criterion as a negative outlier. Further
investigation revealed that all but three of the positive outlier effect sizes
were derived from reports with unusually small samples (i.e., 15 or fewer
youth). Of the remaining three effect sizes, one came from a study that
examined whether providing a mentor to youth who had attended a sum-
mer leadership institute facilitated their completion of a community service
project during the following school year (Mertens, 1988). The outcome mea-
sured in this study, completion of the community service project, was thus
an immediate objective of the program itself. The other two positive effect
sizes appeared as relatively isolated findings within the reports involved.
In summary, although studies that report unusually large effect sizes merit
careful consideration as possible examples of exemplary or “best” practice
(Cooper et al., 2000), there was little evidence to support this interpretation
in the present review. As a safeguard against these extreme values having
undue influence on the findings of subsequent analyses, the effect sizes in-
volved (and two other positive d-indexes that approached criterion as out-
liers) were Winsorized by setting their values to 1.25 or 1.25 in the case
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
168 DuBois, Holloway, Valentine, and Cooper
of the negative d-index outliers. Because the outlier d-index derived from
Mertens (1988) was the only effect size available from this report, the study
continued to represent an extreme observation at the level of independent
sample analysis even after the d-index involved had been Winsorized. On the
basis of this result and the idiosyncratic features of the outcome assessment
involved in this report it was omitted from all subsequent analyses.
The distribution of sample sizes also was examined for extreme values.
Nine independent samples, each appearing in separate reports, met or ap-
proached criteria as statistical outliers. These samples ranged in size from
373 to 47,775 and included those associated with several, recent large-scale
evaluations of mentoring programs such as the national, multisite evaluation
of the effectiveness of BB/BSA (Grossman & Tierney, 1998; Tierney et al.,
1995) in which 959 youth participated. The largest sample size was accounted
for by an evaluation of the Cincinnati Youth Collaborative Mentoring Pro-
gram (Bruce & Mueller, 1994) in which all youth without mentors in the
participating schools served as a comparison group (n = 46,732) for youth
who did receive mentoring (n = 1,043). Because the procedure for weighting
effect sizes was based on sample size, the potential existed for these unusu-
ally large samples to have an overwhelming influence on findings. For this
reason, all nine sample sizes identified as potential outliers were Winsorized
by setting their values to 300.
Overall Effect of Youth Mentoring Programs
After Winsorizing the effect sizes, the average unweighted d-index for
the 574 effect size estimates included for subsequent analysis was d = .18.
Using the 59 independent samples involved as the unit of analysis, the av-
erage unweighted d-index was d = .23. The median effect size was d = .18.
Effect sizes then were evaluated after weighting them by the inverse of their
variance, a procedure that involves differing estimation procedures depend-
ing on whether a fixed-effects or random-effects model is assumed. Under
the fixed-effects assumption, the average effect size for the 59 independent
samples was d = .14. Thus, making no distinctions among effects based on
methodology or program, youth, relationship, or measurement character-
istics, the average youth participating in one of the mentoring programs
included in the present review scored approximately one eighth of a stan-
dard deviation higher in a favorable direction on outcome measures than did
the average youth before or without participation in one of these programs.
The 95% confidence interval for the weighted d-index under the assumption
of fixed effects encompassed a lower value of d = .10 and an upper value
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 169
of d = .18. The practical significance of an effect size also can be expressed
by describing how outcomes for intervention and control groups overlap
(Cooper, 1998, Table 5.3). Using this approach, the average weighted effect
size of d = .14 under the assumption of fixed effects indicates that across dif-
ferent types of programs, the outcome for the average participant in a youth
mentoring program surpassed that of approximately 55% of those in the con-
trol group (i.e., the average youth before or without program participation).
Under the assumption of random effects, the average weighted effect size
estimate increased to d = .18, but encompassed a larger confidence interval
ranging from .11 to .25.
As a check on the robustness of the preceding findings to the “file-
drawer” problem (i.e., lack of publication of studies finding null results),
Fail-Safe-N (FSN) calculations were made (Cooper, 1998). The FSN corre-
sponds to the number of null effects that would have to exist in studies not
included in the meta-analysis to overturn the conclusion that a significant
effect is present. Rosenthal (1979) suggested that FSN be equal to or larger
than five times the number of retrieved studies (or, in the present context,
independent samples) plus 10. Using α = .05 (two-tailed), the FSN for the
present study is 513, a value that substantially exceeds the recommended
resistance number of 305 (59 × 5 + 10). This result, in combination with the
effort that was made to retrieve and include as many unpublished reports as
possible in the review, provides confidence that the overall findings consis-
tent with a positive effect of mentoring programs would not be invalidated
even in the presence of a significant publication bias in the literature.
A stem-and-leaf display of average d-indexes for the 59 independent
samples after individual effect sizes had been Winsorized revealed that 51
of the 59 d-indexes were in the direction of positive effects for youth men-
toring programs (see Table I). Of the remaining eight d-indexes, seven were
in a negative direction and one corresponded exactly to 0. In each of the
former seven cases the negative findings represented half or more of the
findings for independent samples within the evaluation. Illustratively, an
evaluation conducted by the New York City Board of Education (1986)
found declines on measures of school attendance (dsof.07 and .25),
number of courses passed (d =−.93), and grade point average (d =−.78)
for 79 high school students who received the mentoring component of a mul-
ticomponent dropout prevention program. Similarly, youth without a prior
major arrest who participated in the Buddy System program referred to pre-
viously were found to have higher arrest rates in comparison to youngsters in
a randomly assigned control condition (ds =−.27 and .15 at posttest and
2-year follow-up, respectively), although a trend in the opposite direction fa-
voring program youth was found among youth who had been arrested prior
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
170 DuBois, Holloway, Valentine, and Cooper
Table I. Stem-and-Leaf Plot of
Average Effect Sizes for Evalua-
tions of Youth Mentoring Programs
(N = 59 Independent Samples)
Stem Leaf
+1.0 9
+0.9 4
+0.8 3
+0.7 7
+0.6 3479
+0.5 27
+0.4 134689
+0.3 035679
+0.2 2444
+0.1 255667788889
+0.0 01333445567778
0.0 8
0.1
0.2 7621
0.3 8
0.4
0.5 1
to program involvement (ds = .55 and .43, respectively; Fo & O’Donnell,
1975; O’Donnell et al., 1979).
Moderator Analyses of Mentoring Program Effects
Possible moderators of mentoring program effects were investigated us-
ing homogeneity analyses (Cooper & Hedges, 1994; Hedges & Olkin, 1985).
This procedure compares the amount of variance in an observed set of ef-
fect sizes with the amount of variance that would be expected by sampling
error alone (as well as other sources of random influence when assuming a
random effects model); the homogeneity statistic for this type of compari-
son is referred to commonly as Q
b
and it follows a chi-square distribution.
In reporting these statistics in this study, both degrees of freedom and the
number of samples involved, k, will be noted. Whenever feasible, the sig-
nificance of a potential moderator was tested with the moderator treated as
a continuous variable in the homogeneity analysis. This approach was de-
signed to maximize sensitivity in the detection of relevant effects. In several
instances, however, it was necessary to treat moderators as categorical vari-
ables in analyses because of their inherently categorical nature (e.g., type
of outcome measure) or because the degree of variation that was observed
across potential values of the variable was not sufficient to justify treatment
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 171
as a continuous variable. To facilitate interpretation of results
(
Cooper et al.,
2000), in instances in which moderators were tested as continuous variables
average effect sizes are reported for two or more discrete ranges of values
of the variable involved.
An examination of frequency distributions for the coded variables re-
vealed some factors for which there was little or no variation discernible
across studies or samples. Illustratively, with regard to research design, it was
found that in reports comparing two groups of youth (i.e., those receiving
mentoring and those not receiving mentoring) that the youth in the control
or comparison group typically received no intervention (n = 37) as opposed
to some type of intervention other than mentoring (n = 4). Similarly, the
age level of mentors when able to be coded was predominantly in the early
adulthood range (19–29 years old; n = 12), with a small minority (n = 6) in
middle adulthood (30–54) and one in late adulthood (55 or older). Lack of
variation to this extent effectively prohibited reliable analysis of the factors
involved as possible moderators of mentor program effects. There also were
several variables for which certain categories had to be combined in order
to obtain distributions and numbers of categories that would be suitable for
the purpose of moderator analyses. For example, although 44 different types
of outcomes had been distinguished in the original coding, these were col-
lapsed into five more general categories of measures for moderator analyses
(i.e., emotional/psychological well-being, problem or high-risk behavior, so-
cial competence, academic/educational, and career/employment). Finally, in
some instances the information required to code the variable was reported
for only a minority of samples. Of particular note, information regarding the
frequency with which mentors actually had contact with youth in programs
and the amount of time that relationships lasted each were reported for only
12 of the 59 independent samples. Because of the theoretical importance
of these and other selected moderators, they were nonetheless retained for
analysis; clearly, however, the results obtained must be regarded as highly
tentative and exploratory in nature.
Before investigating individual moderators of effect sizes, it is important
to conduct a homogeneity analysis to test whether there is variability in effect
sizes greater than that which would be expected by sampling error around a
single population value (Cooper, 1998). The results of this analysis suggested
that the d-indexes were not all estimating the same underlying population
value, Q(58, k = 59) = 227.70, p <.001, and thus that it was appropriate
to look for characteristics potentially involved in moderating effect size. In
view of the relatively small number of independent samples available for
analysis and limited variability across levels of several potential moderator
variables, those findings that approach but do not reach a conventional level
of statistical significance (i.e., ps <.10) are reported in all tests of individual
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
172 DuBois, Holloway, Valentine, and Cooper
moderators. Caution in the interpretation of findings pertaining to the sig-
nificance of individual moderators is nonetheless certainly warranted given
the large number of tests involved and the associated potential for Type I
error.
Report Information
As shown in Table II, neither year of report nor whether the report was
published (i.e., in a journal article or book) or unpublished (e.g., dissertation)
were significant moderators of effect size. Under the assumption of fixed
effects, however, year of report did approach significance as a moderator,
Q(1, k = 59) = 3.27, p <.10, with a trend evident toward larger effect sizes
for more recent studies (see Table II).
Evaluation Methodology
Neither type of control (pretest–posttest vs. posttest–posttest compar-
ison) nor type of two-group design (random assignment vs. nonequivalent
group) were significant moderators of effect size. However, among stud-
ies employing nonequivalent two-group designs, there was a trend under
the assumption of fixed effects for those that made some attempt to match
youth to report larger effects (d = .20) than those that did not (d = .02),
Q(1, k = 26) = 3.41, p <.10. Among the remaining methodological char-
acteristics of studies that were examined, only sample size was a significant
moderator of effect size. This relation was evident for both fixed effects,
Q(1, k = 59) = 8.90, p <.01, and random effects, Q(1, k = 59) = 8.07,
p <.01. As shown in Table II, this finding reflected a tendency for stud-
ies with smaller sample sizes to report larger program effects.
When investigating possible substantive moderators of effect size in a
meta-analysis, it is recommended that the influence of relevant methodolog-
ical factors be controlled for statistically (Durlak & Lipsey, 1991). In the
present context, these factors included sample size and whether matching
was used for nonequivalent groups, given that each exhibited a significant
or nearly significant association with effect size. As noted, type of control
used as the basis for deriving effect size estimates (i.e., single group pretest–
posttest comparison or two group posttest–posttest comparison) was not
found to be a significant moderator of effect size. Nevertheless, because of
the statistical differences involved in derivation and weighting of each type
of effect size estimate, it was desirable to control for any residual varia-
tion associated with this methodological factor as a possible source of in-
fluence on findings. To implement statistical control for the methodological
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 173
Table II. Moderators of Mentoring Program Evaluation Outcomes
Fixed effects Random effects
Moderator kQ
b
d±95% CI Q
b
d ±95% CI
Report information
Year of report
a
59 3.27
1.07
Prior to 1990 25 .10 .08 .16 .11
1990 or after 34 .16 .05 .19 .05
Type of report 59 1.60 0.19
Unpublished 39 .16 .06 .20 .09
Published 20 .10 .07 .16 .12
Evaluation methodology
Type of control
b
74 0.19 0.00
Pretest–posttest 33 .13 .05 .18 .09
Posttest–posttest 41 .14 .06 .18 .09
Type of two-group design 41 0.52 0.64
Random assignment 15 .12 .10 .12 .11
Nonequivalent group 26 .16 .08 .18 .09
Was nonequivalent group matched? 26 3.41
0.97
No 6 .02 .17 .10 .22
Yes 20 .20 .09 .22 .12
How was d derived? 59 0.07 0.47
Unadjusted means & SDs 52 .14 .05 .19 .08
Adjusted means & SDs 7 .13 .10 .13 .18
Who did evaluation? 57 2.65 0.39
Internal 35 .11 .05 .16 .09
External 22 .19 .08 .21 .12
# of youth in analyses
a
59 8.90
∗∗
8.07
∗∗
<65 30 .25 .09 .26 .12
65 29 .11 .05 .14 .08
Nonactive relationships 55 1.26 0.07
Included 24 .11 .07 .17 .11
Excluded 31 .16 .06 .19 .10
Program features
c
Type of program 59 0.27 0.00
Mentoring alone 38 .16 .06 .16 .09
Multicomponent 21 .14 .07 .17 .10
BB/BSA program? 59 0.05 0.24
Yes 8 .14 .11 .12 .19
No 51 .15 .05 .17 .07
Program goal 59 4.51 3.89
Psychosocial 21 .14 .07 .16 .11
Instrumental 28 .21 .07 .22 .10
Both 10 .08 .10 .05 .15
Geographic location 49 0.72 0.06
Large urban 21 .14 .06 .16 .11
Other 28 .19 .08 .18 .11
Setting for mentoring activities
d
59 7.19
2.66
Community 29 .14 .06 .15 .09
School 16 .07 .11 .11 .14
Workplace 6 .24 .17 .24 .22
Other 8 .28 .13 .27 .18
(Continued )
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
174 DuBois, Holloway, Valentine, and Cooper
Table II. (Continued )
Fixed effects Random effects
Moderator kQ
b
d±95% CI Q
b
d ±95% CI
Mentor compensation 51 0.09 0.02
No 41 .15 .06 .17 .08
Yes 10 .17 .08 .18 .14
Monitoring of implementation
d,e
59 5.59
2.71
No
f
15 .06 .09 .06 .14
Yes 44 .18 .05 .19 .07
Gender of mentors (% male)
a
28 1.12 0.13
Majority female 14 .24 .09 .24 .10
Majority male 5 .13 .10 .12 .15
All male 9 .13 .14 .18 .07
Race/ethnicity of mentors 16 0.24 0.04
(% White)
a
Majority White 10 .21 .09 .22 .14
Majority non-White 6 .21 .13 .22 .19
Mentor background: Helping role/ 50 5.75
3.58
profession
d
No
f
38 .09 .05 .10 .07
Yes 12 .26 .12 .25 .14
Screening of prospective mentors
e
59 2.63 0.21
No
f
31 .11 .07 .15 .10
Yes 28 .18 .06 .18 .09
Matching of mentors and youth
e
59 2.35 0.05
No
f
26 .11 .07 .16 .10
Yes 33 .18 .06 .17 .09
Mentor–youth matching: Gender 33 0.04 0.07
No 12 .19 .11 .16 .14
Yes 21 .18 .06 .18 .09
Mentor–youth matching: Race 33 2.47 1.71
No 26 .15 .06 .15 .08
Yes 7 .26 .11 .26 .15
Mentor–youth matching: Interests 33 0.97 0.57
No 19 .15 .08 .15 .11
Yes 14 .21 .08 .20 .10
Mentor prematch training
e
59 0.22 0.17
No
f
15 .13 .12 .13 .16
Yes 44 .16 .05 .17 .07
Supervision of mentors
e
59 1.99 0.55
No
f
29 .11 .07 .14 .10
Yes 30 .18 .06 .19 .09
Ongoing training of mentors
d,e
59 5.58
4.44
No
f
42 .11 .06 .11 .08
Yes 17 .22 .07 .26 .11
Support groups for mentors
e
59 2.26 1.25
No
f
48 .13 .05 .14 .08
Yes 11 .21 .09 .24 .14
Structured activities for mentors/ 59 6.36
5.54
youth
d,e
No
f
35 .11 .06 .10 .08
Yes 24 .22 .07 .25 .10
(Continued )
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 175
Table II. (Continued)
Fixed effects Random effects
Moderator kQ
b
d±95% CI Q
b
d ±95% CI
Parent support/involvement
d,e
59 9.18
∗∗
3.44
No
f
46 .11 .05 .13 .07
Yes 13 .27 .09 .27 .13
Expectations: Frequency of 59 3.92
0.26
contact
d,e
No
f
18 .08 .08 .14 .12
Yes 41 .18 .05 .18 .08
Expectations: Length of 59 0.02 0.16
relationship
e
No
f
8 .14 .15 .13 .20
Yes 51 .15 .05 .17 .07
Frequency of contact expected
a
41 1.37 0.36
2 hr per week 19 .15 .09 .16 .10
>2 hr per week 22 .20 .07 .19 .09
Length of relationship expected
a
51 1.66 0.35
<12 months 34 .16 .06 .18 .10
12 months 17 .14 .07 .16 .12
Best practices: Theory-based
a
59 12.48
∗∗∗
4.54
<627.04 .08 .07 .10
632.20 .05 .22 .08
Best practices: Empirically 59 20.51
∗∗∗
13.65
∗∗∗
based
a
<436.08 .06 .09 .09
423.24 .07 .25 .09
Characteristics of youth
c,g
Gender (% male)
a
42 0.03 0.02
Majority female 22 .18 .07 .18 .08
Majority male 20 .14 .07 .14 .08
Race/ethnicity (% White)
a
41 0.19 0.21
Majority White 15 .14 .11 .14 .12
Majority non-White 26 .19 .06 .19 .07
Developmental level 41 0.53 0.06
Late childhood/early 20 .17 .06 .17 .11
adolescence
Middle/late adolescence 21 .13 .09 .15 .12
Single-parent family 59 1.61 0.31
No
f
47 .13 .05 .14 .07
Yes 12 .20 .10 .18 .14
Low socioeconomic status 59 2.82
1.07
No
f
37 .11 .06 .12 .08
Yes 22 .19 .07 .19 .09
At-risk status 55 16.20
∗∗
8.73
None 5 .14 .12 .15 .18
Individual 21 .00 .08 .03 .10
Environmental 18 .18 .07 .17 .10
Both 11 .25 .10 .26 .13
Mentor–mentee relationships
c,g
Average frequency of contact
a
12 0.08 0.08
<3 hr per week 7 .17 .13 .17 .17
3 hr per week 5 .20 .09 .20 .16
(Continued )
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
176 DuBois, Holloway, Valentine, and Cooper
Table II. (Continued )
Fixed effects Random effects
Moderator kQ
b
d±95% CI Q
b
d ±95% CI
Average length of relationship
a
12 0.05 0.05
<1 year 6 .23 .11 .23 .11
1 year 6 .14 .13 .14 .13
Assessment of outcomes
c,g
Type of outcome
b
99 4.22 3.19
Emotional/psychological 20 .09 .08 .10 .12
Problem/high-risk behavior 15 .19 .07 .21 .12
Social competence 11 .16 .09 .15 .15
Academic/educational 43 .13 .05 .11 .08
Career/employment 10 .19 .12 .22 .16
Data source
b
82 6.15 1.21
Youth 35 .18 .06 .18 .11
Parent 8 .16 .14 .22 .24
Teacher 7 .25 .14 .21 .23
Administrative records 32 .10 .06 .11 .11
Timing of assessment
b
74 0.66 0.32
During program 24 .12 .07 .13 .10
Immediate posttest 39 .14 .06 .16 .08
Follow-up 11 .10 .09 .12 .14
Length of follow-up
a
11 0.14 0.02
1 year 5 .11 .11 .10 .13
>1 year 6 .09 .13 .10 .15
a
This variable was utilized as a continuous variable in moderator analyses.
b
Individual samples in some instances contributed effect sizes to more than one category of
this potential moderator variable (see discussion of “Unit of Analysis” in text for details);
for this reason, the overall value of k for the test of this variable as a moderator is greater than
the total number of independent samples included in the review (i.e., 59).
c
Analyses of this category of moderator variables includes statistical control for the following
methodological factors: sample size, type of control (i.e. pretest–posttest vs. posttest–posttest),
and whether matching was used for nonequivalent groups.
d
This program feature was included in the empirically based index of best practices.
e
This program feature was included in the theory-based index of best practices.
f
Includes samples from reports in which the moderator was not mentioned.
g
Analyses of this category of moderator variables includes statistical control for theory-based
and empirically based indices of best practices.
p <.10.
p <.05.
∗∗
p <.01.
∗∗∗
p <.001.
factors, all d-index estimates were residualized on sample size, whether
matching was used for nonequivalent groups, and type of control/effect
size estimate (see Cooper et al., 2000, for details). The resulting adjusted
d-indexes were used in all subsequent moderator analyses.
4
4
To investigate the extent to which findings of the remaining moderator analyses were af-
fected by introducing control for relevant methodological features of studies, supplementary
analyses were conducted in which this type of control was not included (i.e., all moderator
analyses reported in Table II relating to program features, characteristics of youth, mentor
mentee relationships, and assessment of outcomes). Results of these analyses were generally
unchanged from those that did include methodological controls with only a limited number
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 177
Program Features
As shown in Table II, whether mentoring was provided alone or as
part of a multicomponent program was not a significant moderator of ef-
fect size. Similarly, neither the comparison of BB/BSA versus non-BB/BSA
programs
5
nor the comparison of programs according to psychosocial versus
instrumental goals yielded significant differences in effect sizes. Effect size
also did not demonstrate a significant relation with geographic program lo-
cation, the setting in which mentoring activities took place, or compensation
of mentors (see Table II). Under the assumption of fixed effects, however,
there was a trend indicating setting for mentoring activities as a moderator,
Q(4, k = 59) = 7.19, p <.10, with lower effect sizes for programs that were
based in schools (d = .07) as opposed to other settings such as the workplace
(d = .24) or community (d = .14). In addition, for the fixed-effect analysis,
monitoring of program implementation was a significant moderator of effect
size, Q(1, k = 59) = 5.59, p <.05, with larger effect sizes found for programs
that reported use of procedures for monitoring implementation (d = .18) in
comparison to those that did not (d = .06). This moderator also approached
significance (p <.10) within the random-effect analysis.
With regard to characteristics of mentors, neither gender nor race/
ethnicity were significant moderators of effect size. Utilization of mentors
with a background in a helping role or profession (e.g., teacher), however,
of exceptions evident in terms of which variables either approached (p <.10) or reached
(p <.05) significance as moderators. Furthermore, these latter variations notwithstanding,
the pattern of effect size estimates across relevant categories or levels of each moderator vari-
able without control for methodological factors was found to be substantively similar to that
which was obtained in primary analyses when this type of control was included (see Table II).
5
It will be recalled that the evaluations of BB/BSA programs included a recent large scale,
multisite investigation of the effectiveness of BB/BSA (Grossman & Tierney, 1998; Tierney
et al., 1995). Using means and standard deviations provided by the study investigators, it
was possible for the present review to compute both preprogram versus postprogram effect
size estimates (i.e., change for youth participating in BB/BSA) and postprogram effect size
estimates (i.e., youth in BB/BSA vs. those in randomized control group at the 18-month follow-
up) for all of the 46 outcome measures included in the research. Average pre–post and post–
post effect size estimates were .02 and .05, respectively, and thus lower than those evident
overall for the eight BB/BSA evaluations included in the present review (ds of .14 and .12
under assumptions of fixed and random effects, respectively). This finding is not necessarily
consistent with the manner in which results of the large-scale evaluation frequently have been
cited by the media (e.g., “Big government,” 1995) and others as demonstrating a large impact
for mentoring relationships. Several factors may be relevant to consider in this regard, however,
including the use of a nonstandard methodology for deriving estimates of the magnitude of
program effects in original reports of the research (e.g., Tierney et al., 1995), the equal weight
given to all outcome measures in the present analysis as opposed to, for example, only those for
which statistically significant effects were found, and, finally, possible enhanced sensitivity to
detecting program influences when incorporating statistical control for variations in baseline
characteristics of study participants as was done in the primary research (Grossman & Tierney,
1998).
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
178 DuBois, Holloway, Valentine, and Cooper
was a significant moderator of effect size under the assumption of fixed-
effects, Q(1, k = 50) = 5.75, p <.05. Evaluations of programs that used
these types of mentors reported larger effect sizes (d = .26) than those for
which utilization of such mentors was not indicated (d = .09). This modera-
tor also approached significance (p <.10) within the random-effect analysis.
The use of procedures for screening prospective mentors was not re-
lated significantly to effect size, nor was matching of mentors and youth
on the basis of relevant criteria. Furthermore, among those programs that
did utilize matching procedures, different types of criteria for matching (i.e.,
gender, race/ethnicity, or interests) were not significant moderators of effect
size.
Although the provision of initial, prerelationship training to mentors
was not related significantly to effect size, a difference was found with re-
gard to the provision of ongoing training during relationships for both fixed
effects, Q(1, k = 59) = 5.58, p <.05, and random effects, Q(1, k = 59) =
4.44, p <.05. Specifically, those programs in which mentors received ongo-
ing training reported larger effects (ds of .22 and .26 for fixed and random
effects, respectively) than those in which this type of training was not indi-
cated to have been made available (ds of .11). As can be seen in Table II,
provision of structured activities for mentors and youth and inclusion of a
parent support or involvement component also were significant moderators
of effect size under the assumption of fixed effects, Q(1, k = 59) = 6.36,
p <.05, and Q(1, k = 59) = 9.18, p <.01, respectively; under the assump-
tion of random effects, provision of structured activities remained a signifi-
cant moderator ( p <.05) and parent support/involvement approached sig-
nificance (p <.10). As can be seen in Table II, evaluations of programs that
included these features reported larger effect sizes than did those of other
programs. Supervision and support groups for mentors were not found to be
significant moderators of effect size.
The final individual program features examined were expectations for
frequency of contact and duration of relationships between mentors and
youth. Expectations regarding frequency of contact was a significant moder-
ator of effect size under fixed effects, Q(1, k = 59) = 3.92, p <.05, with larger
effect sizes reported in evaluations of programs that did include this type of
expectation (d = .18) in comparison to other programs (d = .08). This pro-
gram feature was not a significant moderator, however, when conducting a
random-effect analysis. Expectations regarding the duration of relationships
was not a significant moderator of effect size for the fixed-effect or random-
effect analysis. In addition, among those programs for which either type of
expectation was reported, variations in the frequency of contact or length
of relationship expected were not found to be related significantly to effect
size (see Table II).
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 179
The theory-based index of best practices described previously was
found to be a significant moderator of effect size both under fixed effects,
Q(1, k = 59) = 12.48, p <.001, and random effects, Q(1, k = 59) = 4.54,
p <.05. As shown in Table II, larger effect sizes were reported in evalua-
tions of programs that engaged in a majority of the 11 proposed best practices
(ds of .20 and .22 for fixed and random effects, respectively) in comparison to
other programs (ds of .04 and .07). This indicated cumulative contribution of
theory-based “best practice” indicators to the prediction of greater program
effect sizes and is consistent with the previously described analyses in which
5 of the 11 indicators involved were found to reach or approach significance
as individual moderators of effect size; it is also noteworthy in this regard
that although nonsignificant, variation in average effect size estimates for
each of the remaining six indicators was in the direction of larger effects in
conjunction with the presence of the relevant “best practice” indicator (see
Table II). To further ensure that results for the theory-based index of best
practices did not reflect the influence of a single isolated program feature, a
sensitivity analysis was conducted. Specifically, the index was reexamined as
a moderator in a series of analyses that sequentially excluded each of the 11
theory-based indicators of best practices from the index. Under the assump-
tion of fixed effects, the index remained a significant moderator of effect size
regardless of which indicator was excluded from consideration ( ps <.01).
Similarly, it also generally remained a significant moderator under the as-
sumption of random effects ( ps <.05). The exceptions in the latter instance
were that the finding approached significance (p <.10) only when removing
either ongoing training for mentors or provision of structured activities for
mentors/youth from the index.
As described previously, the empirically based index of best practices
was based on the program features that reached or approached significance
as individual moderators of effect size in the present investigation. A total
of seven program features met this criterion (see Table II).
6
For purposes
of incorporating setting for mentoring activities into the empirically based
best practices index, all settings other than school were considered a “best
practice” (i.e., community, workplace, other) given that consistently larger
effect sizes were found to be associated with these programs relative to
6
Two program features (i.e., setting for mentoring activities and expectations for frequency of
contact) were significant moderators of effect size in the fixed-effect analysis, but did not reach
or approach significance in the random-effect analysis. Nevertheless, to allow for compara-
bility of findings across fixed- and random-effect analyses, a decision was made to compute
a single empirically based best practices index for each study that included consideration of
the presence or absence of both of the program features involved. This relatively inclusive
strategy is consistent with the inherently more conservative nature of random-effect analyses
in general (Wang & Bushman, 1999) as well as their potential to obscure significant moderator
effects when underlying assumptions are not met (Overton, 1998).
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
180 DuBois, Holloway, Valentine, and Cooper
those for which mentoring activities were indicated to have been primarily
school-based. The resulting index was found to be a significant modera-
tor of effect size both under fixed effects, Q(1, k = 59) = 20.51, p <.001,
and random effects, Q(1, k = 59) = 13.65, p <.001. As shown in Table II,
evaluations of programs that engaged in a majority of the relevant prac-
tices reported substantially larger effect sizes (ds of .24 and .25 for fixed
and random effects, respectively) than did other programs (ds of .08 and
.09). A sensitivity analysis conducted using procedures described previously
revealed that this moderating effect was not attributable to any single pro-
gram feature included in the index ( ps <.001 for both fixed and random
effects).
The results of the preceding analyses indicate a trend for greater num-
bers of theory- and empirically based best practice indicators to be associ-
ated with larger effect sizes. They do not address, however, which specific sets
of features included in the best practice indices were most responsible for
these trends. Moderate, but statistically significant zero-order correlations
were evident both among the 11 theory-based best practice indicators (ab-
solute mean r = .18; range from .15 [supervision of mentors and provision
of structured activities for mentors/youth] to .57 [supervision of mentors
and matching of mentors and youth]) and among the 7 empirically based
indicators (mean r = .23; range from .54 [nonschool setting for mentor-
ing activities and background in helping role/profession for mentors] to .37
[monitoring of implementation and background in helping role/profession
for mentors]). Accordingly, a multivariate approach was used to investigate
the extent to which specific indicators comprising each index made inde-
pendent contributions to the prediction of effect size. Specifically, a forward
selection stepwise multiple regression procedure was used to construct a
single best-fitting equation for each type of index. The criterion used for
variable entry was a significant or nearly significant (p <.10) unique contri-
bution to the prediction of effect size independent of other variables already
included in the model; in addition, variables already included as predictors
were eligible for removal at successive steps if their contributions no longer
approached significance. Two alternatives to the squared multiple correla-
tion (i.e., R
2
) have been proposed for quantifying the degree to which a
given set of predictors in a fixed-effects regression model explain or account
for variation in effect sizes. Hedges (1994) recommended use of a descriptive
statistic called the Birge ratio (R
B
); this ratio estimates the ratio of between-
studies variation in effects to the variation due to (within-study) sampling
error. Larger values indicate greater degrees of unexplained variation rela-
tive to a model with exact fit (i.e., Birge ratio of 1); a Birge ratio of 1.5, for
example, suggests that there is 50% more between-studies variation than
might be expected given the within-study sampling variance. Hedges (1994)
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 181
as well as Wang and Bushman (1999) also discussed computing the propor-
tion of “explainable” between-study variation in effect size that is accounted
for by a set of predictors (i.e., the squared multiple correlation divided by
the maximum squared correlation that is possible, taking into account non-
systematic within-study sources of error that are inherently unable to be
accounted for by study-level predictors).
For theory-based best practice indicators, the best-fitting regression un-
der the assumption of fixed effects included parent support/involvement
(b = .13, p <.05) and structured activities for mentors/youth (b = .08,
p <.10) as predictors (R
2
= .12; R
B
= 1.66; 27% of “explainable” between-
study variation). Under the assumption of random effects, the best-fitting
model included structured activities for mentors/youth (b = .14, p <.05)
and ongoing training for mentors (b = .12, p <.10) as predictors.
For empirically based best practice indicators, the best-fitting model
under the assumption of fixed effects included four predictors (R
2
= .24;
R
B
= 1.48; 53% of “explainable” between-study variation): nonschool set-
ting for mentoring activities (b = .26, p <.001), mentor background in a
helping role/profession (b = .22, p <.01), parent support/involvement
(b = .13, p <.05), and structured activities for mentors/youth (b = .08,
p <.10). Under the assumption of random effects, the best-fitting model
included structured activities for mentors/youth and ongoing training for
mentors as predictors and thus was identical to the model identified in the
random-effect analysis for theory-based best practice indicators.
Because of the associations found between program practices and es-
timates of effect size, it was possible that differences in implementation
of relevant practices across studies could introduce bias into analyses of
other types of potential moderator variables. To address this concern, all
d-indexes were adjusted for their associations with both the theory-based
and empirically based indices of best practices, using the same regression
procedure that had been employed previously to control for methodologi-
cal factors as a confounding influence. The resulting d-indexes, now adjusted
for methodological factors and both indices of best practices, were utilized
in the remaining moderator analyses.
7
7
Supplementary analyses for the remaining categories of moderator variables (i.e., characteris-
tics of youth, mentor–mentee relationships, and assessment of outcomes) also were conducted
without the incorporation of control for the theory- and empirically based best practice indices.
With only one exception, results did not differ substantively from those obtained in primary
analyses that did include control for the best practice indices (see Table II). The exception
involved developmental level approaching significance (p <.06) as a moderator of effect size
under the assumption of fixed effects when removing control for the best practice indices (ds
of .21 and .10 for late childhood/early adolescence and middle/late adolescence, respectively),
whereas it did not do so in primary analyses. This difference appears attributable to greater
numbers of both theory- and empirically based best practice indicators for programs in which
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
182 DuBois, Holloway, Valentine, and Cooper
Characteristics of Youth
As shown in Table II, the demographic characteristics of youth and
their families that were examined (i.e., gender, race/ethnicity, developmental
level, single-parent home, socioeconomic status) were not significant moder-
ators of effect size. Under the assumption of fixed effects, however, socioe-
conomic status did approach significance as a moderator, Q(1, k = 59) =
2.82, p <.10. Larger effect sizes were reported for samples of youth from
primarily low socioeconomic backgrounds (d = .19) than for other samples
(d = .11). Consistent with this trend, at-risk status was found to be a signif-
icant moderator of effect size under the assumptions of both fixed effects,
Q(3, k = 55) = 16.20, p <.01, and random effects, Q(3, k = 55) = 8.73,
p <.05. Effect sizes were largest for samples of youth experiencing both
individual and environmental risk factors (ds of .25 and .26 for fixed and
random effects, respectively) or environmental risk alone (ds of .18 and .17).
Average effect size estimates were somewhat lower for the relatively small
number of samples in which youth were not indicated to be experiencing
either type of risk (ds of .14 and .15), with the associated confidence inter-
vals not consistently allowing for the inference of an overall positive effect
of mentoring. Finally, near-zero average estimates of effect size were evi-
dent for those samples of youth indicated to be experiencing individual risk
factors alone (ds of .00 and .03). To further investigate the latter finding,
additional analyses were conducted to determine whether program prac-
tices were related to the magnitude of indicated effects of mentoring on
youth exhibiting individual level risk factors. These analyses sought to ex-
amine whether positive effects of mentoring on these youth might be evident
for those programs that employed relatively greater numbers of the previ-
ously described theory-based and empirically based “best practices. Ac-
cordingly, the estimates of effect size utilized for these supplementary anal-
yses were preadjusted only for the previously noted possible methodological
confounds (i.e., control for number of “best practices” was removed). The
theory-based “best practices” index was found to be a significant moderator
of effect size among the 21 independent samples of youth included within the
individual risk status category under the assumptions of both fixed effects,
Q(1, k = 21) = 14.81, p <.001, and random effects, Q(1, k = 21) = 5.35,
p <.001. Positive effects of mentoring for these youth were evident when
programs engaged in a majority of the relevant practices (ds = .20 and .24
for fixed and random effects, respectively, with 95% confidence intervals of
relatively younger youth received mentoring (i.e., Ms of 6.95 and 4.20 for late childhood/early
adolescent and 4.67 and 2.90 for middle/late adolescent, respectively, ps <.01), such that
larger estimates of effect size at earlier stages of development were evident only when failing
to control for associated variation in relevant features of the programs involved.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 183
±.13 and .18); by contrast, when this was not the case, average effect size esti-
mates were in a negative direction (ds =−.13 and .06, confidence intervals
of .11 and .18, respectively). A majority of the 21 independent samples in-
volved (n = 12) constituted the latter group for which “best practices” were
less evident, thus contributing to the lack of an overall positive effect size for
this category of risk status. Similar results were obtained when examining
the empirically based best practices index as a moderator, for fixed effects:
Q(1, k = 21) = 24.98, p <.001, ds =−.13 and .21, confidence intervals of
.11 and .13, for programs engaging in fewer or more than half of the relevant
practices, respectively; for random effects, Q(1, k = 21) = 11.47, p <.001,
ds =−.07 and .27, confidence intervals of .17 and .22, respectively, with most
samples of youth in the individual-risk status category (n = 13) again partic-
ipating in programs not indicated to be engaging in a majority of the targeted
practices.
Mentor–Youth Relationships
As can be seen in Table II, neither reported average frequency of con-
tact between mentors and youth nor length of relationship was a significant
moderator of effect size. It will be recalled, however, that these analyses
were limited by the small numbers of studies that reported data on either
variable.
Assessment of Outcomes
Type of outcome assessed was not a significant moderator of effect size
(see Table II). Under the assumption of fixed effects, the 95% confidence
intervals associated with effect size estimates were consistent with a positive
effect of mentoring programs on all five types of outcomes examined (i.e.,
emotional/psychological, problem/high-risk behavior, social competence,
academic/educational, and career/employment), although only to a marginal
extent for emotional/psychological adjustment. Under the assumption of
random effects, this was the case for three types of outcomes (i.e., problem/
high-risk behavior, academic/educational, and career/employment), the ex-
ceptions being measures of social competence and emotional/psychological
adjustment.
Similarly, neither data source nor timing of assessment were found to be
significant moderators of effect size. Under the assumption of fixed effects,
confidence intervals for effect size estimates were consistent with favorable
effects of mentoring for all data sources (i.e., youth, parent, teacher, and
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
184 DuBois, Holloway, Valentine, and Cooper
administrative records) and for assessments occurring during programs, at
immediate posttest, and at follow-up. By comparison, under the assumption
of random effects this was the case only when youth constituted the data
source and when assessments took place either during the program or at
an immediate posttest. Length of follow-up assessment was not a significant
moderator in either type of analysis, although the small number of samples
involved precluded inferences of positive effects of mentoring within specific
ranges of this variable (i.e., less than or equal to 1 year vs. greater than 1 year).
Analyses Controlling for At-Risk Status and Type of Outcome
As indicated previously, at-risk status was a significant moderator of
effect size under the assumptions of both fixed- and random-effect anal-
ysis. It therefore was important to consider the extent to which this vari-
able exhibited associations with other moderator variables investigated and
whether or not controlling for these would have any substantial implica-
tions for primary study results. Despite the evidence to suggest that total
number of theory-based or empirically based best practice indicators might
vary significantly across at-risk status category, this was not found to be
the case ( ps >.10). Selected other variables, however, including some of
the indicators that comprised each of these indices, did exhibit significant
covariation with at-risk status. Illustratively, a significant association was ev-
ident between at-risk status of the sample and whether or not mentors had
a background in a helping role or profession, χ
2
(3) = 11.71, p <.01, with
samples of youth in the individual risk status category accounting for a dis-
proportionately large proportion of the instances in which mentors with such
backgrounds were used in programs (i.e., 9 of the 12 independent samples
involved).
To investigate the influence of their associations with at-risk status, all
remaining variables shown in Table II were reevaluated as possible moder-
ators of effect size with statistical control for this characteristic (i.e., resid-
ualizing all effect size estimates on at-risk status of the associated sample,
using three dummy variables to represent the four possible categories of risk
status). Introducing this additional control produced few noteworthy changes
in results. Specifically, with only two exceptions, all variables that had previ-
ously reached or approached significance as moderators in primary analyses
under the assumptions of either a fixed- or random-effects model continued
to do so in these supplementary analyses. The exceptions were that moni-
toring of implementation no longer approached significance as a moderator
under the assumption of random effects, nor did low socioeconomic status
under the assumption of fixed effects.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 185
Further analyses investigated the extent to which results of modera-
tor analyses were robust to possible confounding of the different variables
involved with type of outcome measure utilized in evaluations. As noted
previously, the overall degree of variation in average effect size estimates
across category of outcome measure was not statistically significant. It still
nevertheless was important to examine whether the variation that was ev-
ident represented a source of influence on the findings of other analyses
(cf. Durlak & Wells, 1997). Results of analyses that included control for type
of outcome assessed revealed only a few substantive changes from those
reported in primary analyses. Specifically, as was the case when controlling
at-risk status, low socioeconomic status no longer approached significance
as a moderator under the assumption of fixed effects. Program goal and
screening of prospective mentors also now approached significance as mod-
erators in fixed-effect analyses (ps <.10). These latter findings involved the
same trends that are evident in Table II toward larger estimates of effect
size for those mentoring programs that emphasized instrumental goals for
youth and those that indicated use of procedures for screening prospective
mentors.
Intervention Group Comparisons on Relationship Quality
The final set of analyses investigated effect sizes for comparisons that
were made within the intervention group on the basis of relationship fac-
tors. The information needed to calculate this type of effect size was available
for nine independent samples, each of which appeared in a different study.
The relationship factors assessed in these reports included longevity (Royse,
1998), frequency and amount of contact (Howitt, Moore, & Gaulier, 1998),
and whether or not a mentor was actually received within the context of the
multicomponent Career Beginnings program (Cave & Quint, 1990); in the
remaining studies, broader indices or categories of relationship quality were
derived from sources that included mentor visit reports (Dicken, Bryson, &
Kass, 1977), nominations from teachers (Huisman, 1992) or program staff
(LoSciuto, Rajala, Townsend, & Taylor, 1996), and youth ratings of their
experiences with mentors (Johnson, 1997; Slicker & Palmer, 1993; Stanwyck
& Anson, 1989). Effect sizes were calculated for all relevant comparisons
and coded such that positive values indicated more favorable outcomes for
youth experiencing greater intensity or quality of mentoring. When find-
ings were reported as an association between a continuous relationship
measure and program outcome (e.g., Pearson r ), the finding reported was
converted to a d-index effect size, using the appropriate formula (Cooper,
1998).
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
186 DuBois, Holloway, Valentine, and Cooper
Across the nine independent samples, a total of 35 effect sizes were
able to be calculated for comparisons within the intervention group on the
basis of relationship factors. Following procedures described previously, dis-
tributions of effect size and sample size were inspected for outliers, with
one sample size that qualified as an outlier Winsorized to a less extreme
value (i.e., 300). The resulting average unweighted d-index for the 35 effect
size estimates was d = .22. Using the nine independent samples involved
as the unit of analysis, the average unweighted d-index was .33. When ef-
fect sizes were weighted by the inverse of their variance and a fixed-effects
model was assumed, the average effect size for the nine independent sam-
ples was d = .29. Thus, on average, among youth participating in mentoring
programs, those for whom relationships of greater intensity or quality were
evident scored between one quarter and one third of a standard deviation
higher in a favorable direction on outcome measures. The 95% confidence
interval for this weighted d-index encompassed a lower value of d = .16 and
an upper value of d = .42. Under the assumption of random effects, the av-
erage effect size for the nine independent samples was d = .30 with a 95%
confidence interval extending from d = .15 to d = .45.
DISCUSSION
Findings of this investigation provide support for the effectiveness of
youth mentoring programs. Results of a fixed-effects model analysis indicate
an overall or average positive effect for those specific mentoring programs
that have been the subject of formal evaluation (i.e., those included in the
present review); a random-effects model analysis, furthermore, suggests that
benefits of mentoring may generalize to a broader range of approaches to
implementing this type of intervention. In accordance with the latter find-
ing, moderator analyses revealed little evidence that the potential for pro-
grams to yield desirable outcomes is dependent on such considerations as
whether or not mentoring takes place alone or in conjunction with other
services, whether it is provided in accordance with the most widely imple-
mented model (i.e., BB/BSA), or whether programs reflect relatively general
(i.e., psychosocial) as opposed to more focused (i.e., instrumental) goals. Fa-
vorable effects of mentoring programs are similarly apparent across youth
varying in demographic and background characteristics such as age, gender,
race/ethnicity, and family structure and across differing types of outcomes
that have been assessed using multiple sources of data. Although included in
only a minority of studies, follow-up assessments that have been conducted
also offer at least a limited basis for inferring benefits of mentoring that
extend beyond the end of program participation. Cumulatively, based on
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 187
available findings, it thus seems that youth mentoring programs do indeed
have significant capacity to reproduce through more formal mechanisms the
types of benefits that have been indicated to accrue from so-called natural
mentoring relationships between youth and adults (for reviews, see Rhodes,
1994; Werner, 1995).
Results further indicate, however, that it may be most appropriate to
expect the typical youth participating in a mentoring program to receive
benefits that are quite modest in terms of absolute magnitude. The average
estimated effect sizes of .14 and .18 obtained under the assumptions of fixed
and random effects, respectively, are consistent with only a small effect for
mentoring programs (Cohen, 1988; Lipsey, 1990). This degree of impact,
moreover, falls substantially short of larger mean effect sizes reported pre-
viously for psychological, educational, and behavioral treatments generally
(Lipsey & Wilson, 1993) and for mental health prevention programs directed
at children and adolescents specifically (Durlak & Wells, 1997, 1998). This as-
pect of findings is seemingly inconsistent with the widespread and largely un-
questioned support that mentoring initiatives have enjoyed in recent years.
Nevertheless, strong cautionary views have been offered previously in the
youth mentoring literature (Freedman, 1992; Hamilton & Hamilton, 1992;
Rhodes, 1994). It has been pointed out in particular that numerous pro-
grammatic and other variables may be critical to attend to for the potential
benefits of youth mentoring programs to be fully realized. The need for
greater consideration of specific factors influencing effectiveness is under-
scored by the substantial overall heterogeneity in estimates of effect size
observed in the present review and the numerous systematic sources of this
variation that were able to be delineated in moderator analyses.
Moderators of Program Effectiveness
The theory-based and empirically based indices of best practices for
mentoring programs are particularly noteworthy among the significant mod-
erators of effect size identified. No single feature or characteristic of pro-
grams was indicated to be responsible for the positive trends in outcomes
that were associated with greater degrees of utilization of either set of best
practices. Several of the practices comprising the theory-based index did,
however, emerge as significant individual moderators of effect size (and,
hence, by definition also were included in the empirically based index), thus
highlighting specific strategies that may be especially important for achieving
desired results. These latter program features include ongoing training for
mentors, structured activities for mentors and youth as well as expecta-
tions for frequency of contact, mechanisms for support and involvement of
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
188 DuBois, Holloway, Valentine, and Cooper
parents, and monitoring of overall program implementation. In multivariate
analyses, these practices were further revealed to be represented consis-
tently among the strongest predictors of greater reported positive effects
for mentoring programs. The constellation of program characteristics in-
volved reflects an emphasis on providing adequate support and structure
for mentoring relationships throughout the formative strategies of their de-
velopment (Hamilton & Hamilton, 1992). It is noteworthy therefore that
efforts directed toward this goal apparently have been relatively neglected
in youth mentoring programs to date in lieu of a greater focus on preparatory
procedures such as screening, initial training and orientation, and matching
of youth and mentors. Illustratively, whereas initial training or orientation
has been provided to mentors on a fairly routine basis (71% of studies in
the present review), efforts to provide ongoing training once relationships
have begun have been much less common (23% of studies). Factors such
as increased cost and reluctance to make excessive demands on volunteer
mentors represent potentially formidable obstacles to providing a more sus-
tained infrastructure in programs (Freedman, 1992). Nevertheless, in view
of available findings, it seems clear that at a minimum there is a need for
decision-making in this area to incorporate careful consideration of possible
implications for program outcomes.
A similarly strong linkage with beneficial outcomes is evident for the in-
tensity and quality of relationships established between mentors and youth
in programs. Specifically, among several studies in which comparisons have
been made on the basis of relevant criteria within the intervention group,
a substantial difference on criterion measures is apparent favoring those
youth identified as having relatively strong relationships with their men-
tors. Many of the relationship characteristics reportedly utilized in deriving
such comparisons have been found previously to be predictive of greater
perceived benefits of mentoring as evaluated subjectively by mentors and
youth (DuBois & Neville, 1997; Freedman, 1988; Parra et al., 1998). It ap-
pears based on this research that multiple features of relationships, such
as frequency of contact, emotional closeness, and longevity, each may make
important and distinctive contributions to positive youth outcomes. Unfortu-
nately, it was not feasible to investigate this possibility in the present review
because of the rarity with which measures of specific relationship charac-
teristics have been included in controlled evaluations of mentoring pro-
grams. A related methodological consideration with respect to the relatively
less differentiated appraisals of relationship quality that have been incorpo-
rated into existing evaluation studies is the potential for such judgements
to be contaminated by knowledge of which youth mentees are prospering
most in programs, thus confounding assessments of relationship factors and
outcomes.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 189
A further noteworthy result is the support found for the prevailing view
that mentoring programs offer the greatest potential benefits to youth who
can be considered to be at-risk (Freedman, 1992; Hamilton & Hamilton,
1992). It will be recalled in this regard that the largest estimates of effect
size are evident for programs directed toward youth experiencing condi-
tions of environmental risk or disadvantage, either alone or in combination
with factors constituting individual level risk. A similar trend is apparent
when considering low family socioeconomic status as a specific indicator of
environmental disadvantage. Within the context of frameworks for classi-
fying prevention efforts (Cowen, 1985; Institute of Medicine, 1994), these
findings are consistent with greater effectiveness for mentoring programs
characterized by a situation-focused or selective orientation. Interventions
of this type focus on individuals who can be considered vulnerable by virtue
of their present life circumstances, but who are not yet demonstrating signif-
icant dysfunction. Youth experiencing situations of environmental risk may
be especially suitable candidates for mentoring as a preventive intervention
because of a lack of positive adult support figures or role models in their daily
lives (Rhodes, 1994). With respect to this possibility, available findings do not
indicate reliably greater effects of mentoring for youth from single-parent
households. Enhanced benefits of mentoring have been apparent in the con-
text of low levels of perceived family support (Johnson, 1997), however, thus
suggesting a need for more refined measures of risk associated with the ex-
isting support networks of youth to be included in future research. Exposure
of youth to aspects of environmental adversity not assessed in evaluations
could have additional significance as a factor contributing to the positive
effect of mentoring that was evident to a limited degree even among those
studies for which it was not possible to infer experience of any conditions of
risk on the basis of the information made available.
By contrast, evidence of an overall favorable effect of mentoring is no-
tably lacking under circumstances in which participating youth have been
identified as being at risk solely on the basis of individual-level characteristics
(e.g., academic failure). Mentoring is an inherently interpersonal endeavor.
As a result, it may be especially susceptible to obstacles and difficulties that
can arise when youth targeted for intervention are already demonstrating
significant personal problems (Freedman, 1992). Many of these youth are
likely to be in need of relatively extensive amounts of specialized assistance,
for example, a situation that is not necessarily well-suited to the primarily
volunteer and nonprofessional status of most mentors. Considerations of
this nature suggest a need for training and other appropriate forms of pro-
gram support when attempting to provide effective mentoring to youth who
are exhibiting individual-level risk. In accordance with this view, a more re-
fined analysis revealed that such youth apparently can benefit significantly
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
190 DuBois, Holloway, Valentine, and Cooper
from participation in mentoring programs that adhere to a majority of rec-
ommended practices. Of further note are the substantial positive effects of
mentoring reported for programs in which youth targeted for participation
could be regarded as at-risk from both an individual and environmental per-
spective. Because of the relatively small number of evaluations involved,
this finding merits cautious interpretation. It may be that environmental as
opposed to individual risk simply has greater salience as a determining factor
in likely responsiveness to mentoring. It is also possible, however, that cir-
cumstances of contextual adversity tend to reduce the likelihood of certain
obstacles interfering with efforts to mentor youth who are demonstrating
individual-level risk. In the presence of indications of environmental risk,
for example, mentors may be less prone to accept negative labels assigned
to such youth or inappropriately attribute problems they exhibit solely to
personal deficits or limitations (e.g., lack of motivation).
Applied Implications
From an applied perspective, findings offer support for continued imple-
mentation and dissemination of mentoring programs for youth. The strongest
empirical basis exists for utilizing mentoring as a preventive intervention
with youth whose backgrounds include significant conditions of environ-
mental risk and disadvantage. To facilitate attainment of desired outcomes,
however, results indicate a need for programs to adhere closely to recom-
mended guidelines for effective practice (e.g., National Mentoring Working
Group, 1991). Given the modest size of the effects that thus far have been
able to be established for mentoring, there clearly is a rationale for inno-
vation and experimentation with enhancements to program design. One
possibility suggested by the present findings is the recruitment of mentors
whose backgrounds include prior experience and success in helping roles.
Older adults, for example, although underrepresented currently in programs,
often may be able to bring to the mentoring role valuable skills relating to
child-rearing and other areas of life experience (Freedman, 1988; LoSciuto
et al., 1996). Relative to these needs for both innovation and adherence to
basic guidelines for implementation, concerns such as the most appropri-
ate setting or goals for mentoring activities seem best to regard as being of
secondary importance. Indeed, to the extent that more fundamental consid-
erations are neglected in the development and operation of programs, there
may be substantial opportunity for mentoring to have unintended nega-
tive effects on youth (Rhodes, 1994). This issue seems to warrant particular
attention for those youth who are already exhibiting some degree of personal
vulnerability.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 191
Limitations and Directions for Future Research
Several limitations of the present review also are noteworthy and should
be addressed in future research on youth mentoring programs. One sig-
nificant issue to be kept in mind is that findings do not necessarily reflect
causal effects of either mentoring or the different moderator variables ex-
amined. Positive effects of programs are evident in studies using the most
well-controlled designs (i.e., random assignment) and in those in which men-
toring has been provided alone rather than in combination with other types of
intervention. Yet, even these types of investigations clearly are not immune
to extraneous sources of influence. Consider, for example, the potential that
exists for demand characteristics associated with a youth’s involvement in a
mentoring program to introduce bias into the responses that youth, parents,
and other informants (e.g., teachers) provide on outcome measures. To fully
address this particular concern, it will be important for future evaluations
to more often incorporate “nonreactive” measures into their assessments
of youth outcomes (e.g., archival records of arrests, educational accomplish-
ments, etc.). Given the increasing prevalence of mentoring as an interven-
tion, the possibility that significant numbers of youth within control groups
may themselves be involved in a formal mentoring relationship through in-
volvement in other programs or services also merits greater attention than
it appears to have thus far received in evaluation studies.
Inferences regarding the influence of different moderator variables are
even more tentative because of the inherently correlational nature of any as-
sociations that are found between study characteristics and outcomes within
the framework of meta-analysis (see Cooper, 1998, for further discussion).
Accordingly, priority should be given to more controlled investigation of the
factors identified (e.g., at-risk status) within the context of individual studies
in future research. There also clearly is a related need for evaluations to
more consistently assess characteristics of the relationships that are actually
developed between mentors and youth in programs as a source of influence
on outcomes. These types of efforts, furthermore, should be complemented
by more in-depth consideration of the wide-ranging circumstances within
which mentoring may occur in the life of any given youth.
Issues relating to the generalizability of findings also are a significant
concern. These include possible limitations in the extent to which results
can be extrapolated to the much broader range of mentoring programs
not included in the present review. The importance of this consideration is
underscored by the lack of complete robustness of findings when conducting
analyses under the assumption of a random- rather than fixed-effect model
and by the potential for programs that have not received formal evaluation
to differ systematically from those that have been subjected to this type of
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
192 DuBois, Holloway, Valentine, and Cooper
scrutiny. The ability to make predictions about the efficacy of youth men-
toring programs in the future is similarly prone to uncertainty given the
still evolving status of approaches to intervention in this area. Subsequent
programs, for example, may include significant innovations influencing effec-
tiveness that are not reflected in those programs that have received formal
evaluation to date. Estimates of effect size derived along basic dimensions of
intervention design and evaluation also vary to some extent and thus serve
to illustrate specific areas in which conclusions regarding the effectiveness of
mentoring programs for youth may require qualification. These include po-
tential liabilities associated with restricting mentoring activities to the school
setting, evidence of a relatively weak impact on emotional/psychological
outcomes, and, perhaps most notably, absence of compelling support for
inferring benefits to youth that extend substantially beyond the end of pro-
gram involvement. Cumulatively, the preceding considerations strengthen
the rationale for ongoing evaluation of youth mentoring programs, espe-
cially with respect to those areas for which effectiveness currently is less
well established.
A final recommendation is pragmatic in nature. Because of the diversity
of published and unpublished sources in which mentoring program evalua-
tions have appeared, a great deal of time and effort was required to locate
and obtain the studies included in the present review. Many of these reports
predate earlier reviews, but were not included in them perhaps at least in part
because of similar practical considerations. To facilitate a more orderly and
efficient compilation of mentoring program evaluation data in the future, it is
recommended that a research register be created listing all relevant projects
that are either in progress or completed. The availability of a research reg-
ister has proven helpful in other fields of inquiry (Dickersin, 1994) and in
the mentoring literature could serve a complementary function to the na-
tional data base of programs already in existence (Save the Children, 1999).
Integration of research and practice through such mechanisms offers the best
prospect for future development, evaluation, and dissemination of effective
mentoring programs for youth.
REFERENCES
References marked with an asterisk indicate studies included in the meta-analysis.
Abbott, D. A., Meredith, W. H., Self-Kelly, R., & Davis, M. E. (1997). The influence of a
Big Brothers program on the adjustment of boys in single-parent families. Journal of
Psychology, 131, 143–156.
Abcug, L. (1991). Teachers Achieving Success With Kids (TASK): A teacher–student mentor-
ship program for at-risk students (M. S. Practicum, Nova University). (ERIC Document
Reproduction Service No. ED330974)
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 193
Aiello, H. S. (1988). Assessment of a mentor program on self-concept and achievement
variables of middle school underachievers (Doctoral dissertation, Virginia Polytech-
nic Institute and State University, 1988). Dissertation Abstracts International, 49(07),
1699A.
Baldwyn Separate School District, MS. (1982). A gifted model designed for gifted students in a
small, rural high school. Washington, DC: Office of Elementary and Secondary Education.
(ERIC Document Reproduction Service No. ED233847)
Banta, T. W., & Lawson, S. S. (1980). Evaluation of the Lenoir City Schools (Tennessee) Re-
tirement Power in Education Project, 1979–1980. Knoxville, TN: Bureau of Educational
Research and Service. (ERIC Document Reproduction Service No. ED201044)
Big government, big brother. (1995, December 25). The New Republic,7.
Blakely, C. H., Menon, R., & Jones, D. C. (1995). Project BELONG: Final report. College
Station, TX: Texas A&M University, Public Policy Research Institute.
Brooks, L. J. (1995). An evaluation of the VCU Mentoring Program (Doctoral thesis, Virginia
Commonwealth University, 1995). Dissertation Abstracts International, 57(04), 1481A.
Bruce, D., & Mueller, E. J. (1994). Mentoring: A Cincinnati Youth Collaborative program.
Evaluation of outcomes 1993–1994. Cincinnati, OH: Cincinnati Youth Collaborative.
Buman, B., & Cain, R. (1991). The impact of short term, work oriented mentoring on the em-
ployability of low-income youth. (Available from Minneapolis Employment and Training
Program, 510 Public Service Center, 250 S. 4th St., Minneapolis, MN 55415)
Cave, G., & Quint, J. (1990). Career beginnings impact evaluation: Findings from a program
for disadvantaged high school students. New York: Manpower Demonstration Research
Corporation. (ERIC Document Reproduction Service No. ED325598)
Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ: Erlbaum.
Colson, S. (1979). Evaluating career education for gifted and talented students. Journal of
Research and Development in Education, 12, 51–62.
Cooper, H. (1998). Synthesizing research: A guide for literature reviews (3rd ed.). Thousand
Oaks, CA: Sage.
Cooper, H., Charlton, K., Valentine, J. C., & Muhlenbruck, L. (2000). Making the most of sum-
mer school: A meta-analytic and narrative review. Monographs of the Society for Research
in Child Development, 65(1, Serial No. 260).
Cooper, H., & Hedges, L. V. (Eds.). (1994). Handbook of research synthesis. New York: Russell
Sage Foundation.
Cowen, E. L. (1985). Person-centered approaches to primary prevention in mental health:
Situation-focused and competence-enhancement. American Journal of Community Psy-
chology, 13, 31–48.
Cowen, E. L., Hightower, A. D., Pedro-Carroll, J., Work, W. C., Wyman, P. A., & Haffey, W. C.
(1996). School based prevention for at-risk children: The Primary Mental Health Project.
Washington, DC: American Psychological Association.
Darling, N., Hamilton, S. F., & Niego, S. (1994). Adolescents’ relations with adults outside the
family. In R. Montemayor & G. R. Adams (Eds.), Personal relationships during adolescence
(pp. 216–235). Thousand Oaks, CA: Sage.
Davidson, W. S., & Redner, R. (1988). The prevention of juvenile delinquency: Diversion from
the juvenile justice system. In R. H. Price, E. L. Cowen, R. P. Lorion, & J. Ramos-McKay
(Eds.), Fourteen ounces of prevention: Theory, research, and prevention (pp. 123–137). New
York: Pergamon.
Davis, H. (1988). A mentor program to assist in increasing academic achievement and at-
tendance of at-risk ninth grade students (Doctoral thesis, University of Pittsburgh, 1988).
Dissertation Abstracts International, 50(03), 0580A.
Davison, A. R. (1994). A culturally-enriched career and tutorial program and the vocational
interests and career beliefs of at-risk middle school students (Doctoral dissertation, Lehigh
University, 1994). Dissertation Abstracts International, 56(02), 492A.
Dicken, C., Bryson, R., & Kass, N. (1977). Companionship therapy: A replication of ex-
perimental community psychology. Journal of Consulting and Clinical Psychology, 4,
637–642.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
194 DuBois, Holloway, Valentine, and Cooper
Dickersin, K. (1994). Research registers. In H. Cooper & L. V. Hedges (Eds.), Handbook of
research synthesis (pp. 71–83). New York: Russell Sage Foundation.
DuBois, D. L., & Neville, H. A. (1997). Youth mentoring: Investigation of relationship charac-
teristics and perceived benefits. Journal of Community Psychology, 25, 227–234.
Durlak, J. A., & Lipsey, M. W. (1991). A practitioner’s guide to meta-analysis. American Journal
of Community Psychology, 19, 291–332.
Durlak, J. A., & Wells, A. M. (1997). Primary prevention mental health programs for children
and adolescents. American Journal of Community Psychology, 25, 115–152.
Durlak, J. A., & Wells, A. M. (1998). Evaluation of indicated preventive intervention (secondary
prevention) for children and adolescents. American Journal of Community Psychology, 26,
775–802.
Flaherty, B. P. (1985). An experiment in mentoring high school students assigned to basic
courses (Doctoral thesis, Boston University, 1985). Dissertation Abstracts International,
46(02), 0352A.
Flaxman, E., Ascher, C., & Harrington, C. (1988). Youth mentoring: Programs and practices
(Urban Diversity Series No. 97). New York: Teachers College, Columbia University. (ERIC
Document Reproduction No. ED308257)
Fo, W. S., & O’Donnell, C. R. (1974). The Buddy System: Relationship and contingency con-
ditions in a community intervention program with nonprofessionals as behavior change
agents. Journal of Consulting and Clinical Psychology, 42, 163–169.
Fo, W. S., & O’Donnell, C. R. (1975). The Buddy System: Effects of community intervention
on delinquent offenses. Behavior Therapy, 6, 522–524.
Freedman, M. (1988). Partners in growth: Elder mentors and at-risk youth. Philadelphia: Private/
Public Ventures.
Freedman, M. (1992). The kindness of strangers: Reflections on the mentoring movement.
Philadelphia: Public/Private Ventures.
Galvin, P. F. (1989). Concept mapping for planning and evaluation of a big brother/big sister
program. Evaluation and Program Planning, 12, 53–57.
George, A. Z. (1986). Big Brothers of Greater Los Angeles: An evaluation from a social
learning theory approach (Doctoral thesis, California School of Professional Psychology,
Los Angeles). Dissertation Abstracts International, 46(08), 2806B.
Goodman, S. (1972). Companionship therapy: Studies in structured intimacy. San Francisco:
Jossey-Bass.
Graber, M. (1985, March). The effects of an internship program on the psychosocial devel-
opment of high school students. Paper presented at the annual meeting of the American
Educational Research Association, Chicago, IL.
Green, B. C. (1979). An evaluation of a Big Brothers’ program for father-absent boys: An eco-
behavioral analysis (Doctoral thesis, New York University, 1979). Dissertation Abstracts
International, 41(02), 0671B.
Grossman, J. B., & Tierney, J. P. (1998). Does mentoring work? An impact study of the Big
Brothers Big Sisters program. Evaluation Review, 22, 403–426.
Hamilton, S. F., & Hamilton, M. A. (1992, March). Mentoring programs: Promise and paradox.
Phi Delta Kappan, 546–550.
Harmon, M. A. (1995). Reducing drug use among pregnant and parenting teens: A program
evaluation and theoretical examination (Doctoral thesis, University of Maryland College
Park, 1995). Dissertation Abstracts International, 56(08), 3319A.
Hayes, G. L. (1998). An evaluation of a staff mentor program for at-risk students in an
Oregon high school: CAKE (Caring About Kids Effectively) (Doctoral thesis, Portland
State University, 1998). Dissertation Abstracts International, 59(05), 1517A.
Hayward, B. J., & Tallmadge, G. K. (1995). Strategies for keeping kids in school: Evaluation
of dropout prevention and reentry projects in vocational education (Final Report).
Washington, DC: American Institutes for Research in the Behavioral Sciences. (ERIC
Document Reproduction Service No. ED385767)
Hedges, L. V. (1994). Fixed effects models. In H. Cooper & L. V. Hedges (Eds.), Handbook of
research synthesis (pp. 285–299). New York: Russell Sage Foundation.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 195
Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. Orlando, FL: Academic
Press.
Hedges, L. V., & Vevea, J. L. (1998). Fixed and random effects models in meta-analysis. Psy-
chological Methods, 3, 486–504.
Hill, M. J. (1972). Partners: Community volunteer and probationer in a one-to-one relationship.
Juneau, AL: Department of Health and Social Services, Division of Corrections.
Howitt, P. S., Moore, E. A., & Gaulier, B. (1998). Winning the battles and the wars: An
evaluation of comprehensive, community-based delinquency prevention programming.
Juvenile and Family Court Journal, 49, 39–49.
Huisman, C. (1992). Student Mentoring Program 1989–1992: Evaluation report. Portland, OR:
Oregon Community Foundation. (ERIC Document Reproduction Service No. ED356701)
Institute of Medicine. (1994). Reducing risks for mental disorders. Washington, DC: National
Academy Press.
Johnson, A. W. (1997). Mentoring at-risk youth: A research review and evaluation of the im-
pacts of the Sponsor–A-Scholar Program on student performance (Doctoral dissertation,
University of Pennsylvania, 1997). Dissertation Abstracts International, 58(03), 813A.
Johnson, A. W., & Sullivan, J. W. (1995). Mentoring program practices and effectiveness. In
M. Galbraith & N. Cohen (Eds.), Mentoring: New strategies and challenges (pp. 43–56).
San Francisco: Jossey-Bass.
Joseph, J. A. (1992). Improving self-esteem of at-risk students. (Educational Specialist
Practicum, Nova University). (ERIC Document Reproduction Service No. ED343106)
Keenan, J. T. (1992). Evaluation of a Big Brothers/Big Sisters program for children of alcoholics
(Doctoral thesis, University of Pennsylvania, 1992). Dissertation Abstracts International,
53(07), 2249A.
Lakes, K. D. (1997). The effects of a high school mentoring program on selected academic
variables. Unpublished master’s thesis, Wright State University, Dayton, OH.
Laughrey, M. C. (1990). The design and implementation of a mentor program to improve the
academic achievement of Black male high school students (Educational Specialist Practicum
Report, Nova University). (ERIC Document Reproduction Service No. ED328647)
Lee, S., Plionis, E., & Luppino, J. (1989). Keep Youth in School: A community based practice
model to keep at risk youth in school: Final report. Washington, DC: Catholic University
of America. (ERIC Document Reproduction Service No. ED314676)
Lipsey, M. W. (1990). Design sensitivity: Statistical power for experimental research. Newbury
Park, CA: Sage.
Lipsey, M. W., & Wilson, D. B. (1993). The efficacy of psychological, educational, and behavioral
treatment: Confirmation from meta-analysis. American Psychologist, 48, 1181–1209.
LoSciuto, L., Rajala, A. K., Townsend, T. N., & Taylor, A. S. (1996). Outcome evaluation of
Across Ages: An intergenerational mentoring approach to drug prevention. Journal of
Adolescent Research, 11, 116–129.
McCord, J. (1992). The Cambridge–Somerville study: A pioneering longitudinal experimental
study of delinquency prevention. In J. McCord & R. J. Tremblay, (Eds.), Preventing an-
tisocial behavior: Interventions from birth through adolescence (pp. 196–206). New York:
Guilford.
McLearn, K. T., Colasanto, D., & Schoen, C. (1998). Mentoring makes a difference: Findings
from The Commonwealth Fund 1998 Survey of Adults Mentoring Young People. New York:
The Commonwealth Fund.
McPartland, J. M., & Nettles, S. (1991). Using community adults as advocates or mentors for
at-risk middle school students. American Journal of Education, 99, 637–642.
Mertens, B. L. (1988). An evaluation of the Governors’ School Community Leadership Projects
and Civic Mentor Network. (Doctoral Dissertation, Seattle University, 1988). Dissertation
Abstracts International, 49(08), 2169A.
Mitchell, H., & Casto, G. (1988). Team education for adolescent mothers. Logan, UT: Early
Intervention Research Institute. (ERIC Document Reproduction Service No. ED299744)
National Mentoring Working Group. (1991). Mentoring: Elements of effective practice.
Washington, DC: National Mentoring Partnership.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
196 DuBois, Holloway, Valentine, and Cooper
Nelson, C., & Valliant, P. M. (1993). Personality dynamics of adolescent boys where the father
was absent. Perceptual and Motor Skills, 76, 435–443.
Nettles, S. M. (1991). Community contributions to school outcomes of African-American stu-
dents. Education and Urban Society, 24, 41–52.
New York City Board of Education. (1986). High School Attendance Improvement/
Dropout Prevention Program 1984–1985: Final report (OEA Evaluation Report).
Brooklyn: Office of Educational Assessment. (ERIC Document Service Reproduction
No. ED271529)
O’Donnell, C. R., Lydgate, T., & Fo, W. S. (1979). The Buddy System: Review and follow-up.
Child Behavior Therapy, 1, 161–169.
Overton, R. C. (1998). A comparison of fixed-effects and mixed (random-effects) mod-
els for meta-analysis tests of moderator variable effects. Psychological Methods, 3,
354–379.
Parra, G. R., DuBois, D. L., Neville, H. A., Nanda, S., Mosakowski, K., Sainz, E., et al. (1998,
August). Mentoring relationships for youth: Predictors of perceived benefits and longevity.
Paper presented at the 106th annual convention of the American Psychological Associa-
tion, San Francisco.
Powers, L. E., Sowers, J., & Stevens, T. (1995). An exploratory randomized study of the impact
of mentoring on the self-efficacy and community-based knowledge of adolescents with
severe physical challenges. Journal of Rehabilitation, 61(1), 33–41.
Powers, S., & McConner, S. (1997). Project SOAR 1996–1997: Evaluation report. Tucson, AZ:
Creative Research Associates. (ERIC Document Reproduction Service No. ED412269)
Quint, J. (1991). Project redirection: Making and measuring a difference. Evaluation and
Program Planning, 14, 75–86.
Reller, D. J. (1987). A longitudinal study of the graduates of the Peninsula Academies: Final
report. Palo Alto, CA: American Institutes for Research in the Behavioral Sciences. (ERIC
Document Reproduction Service No. ED327601)
Rhodes, J. E. (1994). Older and wiser: Mentoring relationships in childhood and adolescence.
Journal of Primary Prevention, 14, 187–196.
Rhodes, J. E., Haight, W. L., & Briggs, E. C. (1999). The influence of mentoring on the peer
relationships of foster youth in relative and non-relative care. Journal of Research on
Adolescence, 9, 185–201.
Rippner, M. (1992). Using businesses as on-site-schools to increase academic achievement and
develop employability skills of at-risk students (M.S. Assignment, Nova University). (ERIC
Document Reproduction Service No. ED352459)
Roberts, A., & Cotton, L. (1994). Note on assessing a mentor program. Psychological Reports,
75, 1369–1370.
Rosenthal, R. (1979). Replications and their relative utility. Replications in Social Psychology,
1, 15–23.
Rosenthal, R. (1994). Parametric measures of effect size. In H. Cooper & L. V. Hedges
(Eds.), Handbook of research synthesis (pp. 231–244). New York: Russell Sage
Foundation.
Rowland, R. G. (1992). An evaluation of the effects of a mentoring program on at-risk students
in selected elementary schools in the North East Independent School District (Doctoral
dissertation, Texas A&M University, 1991). Dissertation Abstracts International, 53(1),
0039A.
Royse, D. (1998). Mentoring high-risk minority youth: Evaluation of the Brothers Project.
Adolescence, 33, 145–158.
Saito, R. N., & Blyth, D. A. (1992). Understanding mentoring relationships. Minneapolis, MN:
Search Institute.
Save the Children. (1999). Overview of Save the Children’s National Mentoring Database
[On-line]. Available: www.savethechildren.org/mentors2d.html
Seidle, F. W. (1982). Big Sisters: An experimental evaluation. Adolescence, 17, 117–128.
Slicker, E. K., & Palmer, D. J. (1993). Mentoring at-risk high school students: Evaluation of a
school-based program. The School Counselor, 40, 327–333.
P1: GVH/HAA/GXN/HGA P2: GVH
American Journal of Community Psycgology [ajcp] PP402-368415 February 27, 2002 21:20 Style file version Nov. 19th, 1999
Effectiveness of Mentoring Programs 197
Smith, M. A. B. (1990). The teen incentive program: A research and evaluation model for
adolescent pregnancy prevention (Doctoral dissertation, Columbia University, 1990). Dis-
sertation Abstracts International, 51(09), 3244A.
Stanwyck, D. A., & Anson, C. A. (1989). The Adopt-A-Student Evaluation Project: Final report.
Atlanta: Georgia State University, Department of Educational Foundations.
Tierney, J. P., Grossman, J. B., & Resch, N. L. (1995). Making a difference. An impact study of
Big Brothers/Big Sisters. Philadelphia: Public/Private Ventures.
Tukey, J. W. (1977). Exploratory data analysis. Reading, MA: Addison-Wesley.
Turner, S., & Scherman, A. (1996). Big brothers: Impact on little brothers’ self-concepts and
behaviors. Adolescence, 31, 875–882.
Valenzuela-Smith, M. (1984). The effectiveness of a tutoring program for junior high Latino
students (Doctoral thesis, University of San Francisco, 1984). Dissertation Abstracts Inter-
national, 46(03), 0634A.
Wang, M. C., & Bushman, B. J. (1999). Integrating results through meta-analytic review using
SAS Software. Cary, NC: SAS Institute.
Watson, S. H. (1996). A study on the effects of four different types of mentoring plans using
retired community individuals and college students as mentors on Hispanic “at risk” stu-
dents’ grades and attendance (Masters thesis, Stephen F. Austin State University, 1996).
Masters Abstracts International, 35(05), 1133.
Werner, E. E. (1995). Resilience in development. Current Directions in Psychological Science,
4, 81–85.
... Dozens more published studies have considered qualities of mentors and programs, specific youth populations, and how to best support young people through mentoring interventions. This ever-growing field of study has, so far, established that there are small-to-moderate, short-term positive outcomes for young people [1,2]. Positive outcomes for young people are more likely to occur, with greater effect size, when programs are of high quality [2]. ...
... This ever-growing field of study has, so far, established that there are small-to-moderate, short-term positive outcomes for young people [1,2]. Positive outcomes for young people are more likely to occur, with greater effect size, when programs are of high quality [2]. ...
Article
Full-text available
Building on previous work examining the three central theoretical models driving the youth mentoring literature, the present paper presents an updated conceptual framework on how youth mentoring can equitably support health outcomes for young people, particularly minoritized or otherwise marginalized youth. Youth mentoring has been demonstrated to support positive health outcomes (e.g., mental health, well-being) for all young people, and has a growing literature base to match the enthusiasm in findings. The core conceptual models, however, had not been updated for nearly 20 years. This paper starts with the guiding values behind the updated model, including centering the pursuit of social justice, a recognition of structural oppression, and utilizing key modern theoretical bases (healing-centered engagement, a strengths-based approach, and community cultural wealth). Ultimately, this paper presents an updated conceptual model, outlining key aspects needed to support mental health for minoritized young people through youth mentoring, including building a foundational relationship, key mechanisms of mentoring, reciprocal benefits, and context-specific support.
... Provide additional educational resources on the Internet. Research by DuBois et al. showed that frequent exchanges between mentors and youth mentors doubled the effectiveness of the mentorship program [22]. Therefore, Facebook can create exchange opportunities for discussions between teachers and students. ...
Article
Full-text available
This study analyzes the development of writing skills through the pedagogical use of motivational stories and videos on Facebook. Participants in this study were students from different parts of the country. Particularly, this mixed-methods research aims to determine how these technological tools help students enhance their writing skills at the same time find out students’ perceptions about the use of Facebook. In order to achieve these goals, a Facebook group page was created. The students were then asked to complete different activities, like reading short motivational stories and watching several motivational videos. The intervention lasted for two months. The students’ progress was assessed using rubric for measuring their ability to discover the main idea of the stories and videos, reporting the event sequence, using correct grammar and vocabulary, and summarizing information. A survey was conducted to determine students’ perceptions about the use of Facebook in this particular context. The results show that the pedagogical use of Facebook was positive since it catches students’ attention and helps them develop writing activities collaboratively. Keywords: writing, reading, watching, skill, Facebook, pedagogical. Resumen Este estudio analiza el desarrollo de las habilidades de escritura mediante el uso pedagógico de historias y videos usando Facebook. Este estudio se realizó con estudiantes de diferentes partes del país. Particularmente, el propósito de esta investigación es determinar cómo estas herramientas tecnológicas ayudan a los estudiantes a mejorar su escritura y, correlativamente, conocer las percepciones de los estudiantes sobre el uso de Facebook. Para lograrlo, se creó un grupo en Facebook. Se pidió a los estudiantes que completaran diferentes actividades, como leer historias y ver videos de motivación. La intervención duró dos meses. El progreso de los estudiantes se evaluó con una rúbrica para medir su capacidad para descubrir la idea principal de las historias y videos, informar la secuencia de eventos, usar la gramática y el vocabulario correctos y resumir la información. Se aplicó una encuesta para conocer las percepciones de los estudiantes sobre el uso de Facebook en este contexto. Los resultados muestran que el uso pedagógico de Facebook es positivo ya que llama la atención de los estudiantes para desarrollar actividades de escritura de forma colaborativa. Palabras Clave: escritura, lectura, observación, habilidad, Facebook, pedagógico.
... Thus, research was reviewed to assess the degree to which appropriate programs are available to provide these supports. The review focuses on mentoring programs given (a) their ubiquity in both community and school settings, and (b) their track record in promoting positive behavioral, social-emotional, and academic outcomes for youth of various ages and race/ethnicities (see reviews by DuBois et al., 2002;DuBois et al., 2011;Sánchez et al., 2018). Mentoring programs focus on the development of a relationship between mentor and mentee characterized by mutuality, trust, and empathy. ...
Article
Full-text available
Introduction Adolescent males with disabilities face unique challenges, and mentoring programs designed for this population could support more positive long‐term outcomes. In the current study, a scoping review of empirical research on such programs was conducted. The review was intended to capture the characteristics of existing mentoring program for males with disabilities and map those characteristics in a way that sheds light on the overall status of the field. Methods The review included different types of mentoring (e.g., adult to child, peer to peer, etc.) as long as the program explicitly defined the formation of a long‐term relationship between mentor and mentee. The search identified 21 relevant studies that were categorized using four dimensions: 1) Setting (i.e., school/after school, community‐based, on‐line/virtual, or mixed); 2) Sex (i.e., males only or mixed males/females); 3) Outcomes (i.e., academics, social‐emotional skills, health, transition from high school, or program implementation); and 4) Evaluation methodology (i.e., experimental, nonexperimental, or qualitative). The findings are summarized in an Evidence Gap Map. Results Only three studies used an experimental design, although they were able to demonstrate significant promise in promoting positive outcomes for youth with disabilities. Beyond these studies, however, there was little evidence supporting program efficacy for males with disabilities, as most studies in the sample were nonexperimental in nature. Conclusions Overall there are few mentoring programs that target this population, and experimental research on programs that target academic and health outcomes is particularly lacking.
Chapter
The role of mentoring in countering entrepreneurial loneliness, a commonplace problem in the business world, is analyzed in this chapter. It discusses how the entrepreneurial resilience of boosting emotional security, practical advice, and networking opportunities is provided by mentoring. The adoption of mentorship into the digital realm via virtual platforms, matching through AI, and peer-to-peer mentorship are discussed in the chapter. While discussing conflicts such as mismatched expectations and constraints in terms of time, the chapter also throws up the transformational effects that mentorship has on entrepreneurs geared towards helping them through various stages. As the entrepreneurship landscape continues to change, then creating a successful business or individual development, positive mentorship remains a crucial enabler that empowers entrepreneurs to cut across all the complexities of their ventures with confidence.
Article
Guided by self-determination theory and the bioecological model, this latent class analysis explores the patterns and predictors of supportive relationships in a national sample of early adolescents ( N = 6,469). A six-class solution emerged: youth with (1) emotional, informational, and academic support from various adults and peers across home, school, and community settings; (2) emotional and informational support from school friends; (3) low support; (4) emotional, informational, and academic support from parents and school friends; (5) emotional and informational support from out of school friends, and (6) emotional and informational support from family. Multinomial logistic regression results indicate the unequal distribution of relational support was predicted by individual, home, or school characteristics: being female, extracurricular participation, and higher socioeconomic home or school contexts were associated with membership in classes with a pattern of greater relational support in comparison to the low support class. The importance of targeting subgroups of adolescents in need of relational support and enhancing home and school settings are discussed.
Chapter
School nutrition programs reach approximately 30 million American children each year and provide an ideal opportunity to improve child dietary behaviors, including food waste reduction. Yet, school nutrition programs are challenging to implement and can be difficult research settings. As such, many schools seek low-cost strategies to improve diet quality and reduce food waste. This chapter explores the role of community partnerships, using Cooperative Extension as a case study, to implement nudge interventions to improve fruit and vegetable consumption while reducing waste. In this mixed-methods study, plate waste outcomes were used to evaluate the efficacy of the nudge interventions; interviews with school nutrition staff were used to assess the implementation of the intervention that was delivered in partnership with Cooperative Extension. Nutrition services staff viewed their partnerships with Cooperative Extension as a key facilitator in implementing the interventions and reportedly may not have implemented the nudges without this collaboration. These findings underscore the importance of community partnerships to facilitate innovation in school nutrition programs.
Article
Full-text available
Conventional reviews of research on the efficacy of psychological, educational, and behavioral treatments often find considerable variation in outcome among studies and, as a consequence, fail to reach firm conclusions about the overall effectiveness of the interventions in question. In contrast meta-analytic reviews show a strong, dramatic pattern of positive overall effects that cannot readily be explained as artifacts of meta-analytic technique or generalized placebo effects. Moreover, the effects are not so small that they can be dismissed as lacking practical or clinical significance. Although meta-analysis has limitations, there are good reasons to believe that its results are more credible than those of conventional reviews and to conclude that well-developed psychological, educational, and behavioral treatment is generally efficacious.
Article
Full-text available
Disadvantaged 6–13 yr old Caucasian children were assigned to a Caucasian college student (21 male and 22 female dyads) or to a control group in a replication of G. Goodman's (1972) study of companionship therapy. The student's therapeutic talent was assessed by the Group Assessment of Interpersonal Traits (GAIT). Companioned children improved significantly more than controls on 3 parent-rated problem measures, and there was a trend for greater improvement of self-concept. Teachers perceived little change in either group. Most GAIT variables were not predictively valid, but empathy predicted relationship quality. The pattern of Goodman's findings for dyad types was repeated in the parent data: Quiet or outgoing Ss with an outgoing student improved most, and Ss in a double-quiet dyad improved least. There were relatively few sex differences.
Article