ArticlePDF Available

Over three decades of longitudinal research on the development of foster children: A meta-analysis

Authors:
  • Jeugdbescherming west

Abstract and Figures

Large numbers of children over the world experience foster care each year. How best to satisfy their developmental needs and how to avoid placement breakdowns and negative consequences of foster care are important challenges. In this study, a series of four meta-analyses is performed to examine the longitudinal developmental outcomes of children in foster care. The focus is on adaptive functioning and behavioral outcomes. A literature search identified 11 studies suitable for inclusion in the meta-analysis on adaptive functioning (N=1,550), 24 studies for the meta-analysis on internalizing problems (N=1,984), 21 studies for the meta-analysis on externalizing problems (N=1,729) and 25 studies for the meta-analysis on total behavior problems (N=2,523). No overall improvement or deterioration was found for adaptive functioning. However, studies with a timespan longer than one year and studies with larger sample sizes showed development toward more negative adaptive functioning than studies with shorter timespans or smaller samples. No overall increases or decreases in internalizing, externalizing or total behavior problems were found. Based on these results, it is concluded that foster care does not negatively or positively affect foster children's developmental trajectories. Given that many children enter foster care with problems, this is a worrying situation. Further longitudinal research to find the factors necessary for improving foster children's developmental chances is recommended. Furthermore, routine screening and targeted foster-care interventions are adviseable to ensure that all children, who cannot be raised by their own parents, receive the support conducive to their positive development. Copyright © 2015 Elsevier Ltd. All rights reserved.
Content may be subject to copyright.
Child
Abuse
&
Neglect
42
(2015)
121–134
Contents
lists
available
at
ScienceDirect
Child
Abuse
&
Neglect
Research
article
Over
three
decades
of
longitudinal
research
on
the
development
of
foster
children:
A
meta-analysis
Anouk
Goemans,
Mitch
van
Geel,
Paul
Vedder
Institute
of
Education
and
Child
Studies,
Leiden
University,
Wassenaarseweg
52,
2333
AK
Leiden,
The
Netherlands
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
11
December
2014
Received
in
revised
form
28
January
2015
Accepted
2
February
2015
Available
online
25
February
2015
Keywords:
Meta-analysis
Foster
care
Child
development
Longitudinal
a
b
s
t
r
a
c
t
Large
numbers
of
children
over
the
world
experience
foster
care
each
year.
How
best
to
satisfy
their
developmental
needs
and
how
to
avoid
placement
breakdowns
and
negative
consequences
of
foster
care
are
important
challenges.
In
this
study,
a
series
of
four
meta-
analyses
is
performed
to
examine
the
longitudinal
developmental
outcomes
of
children
in
foster
care.
The
focus
is
on
adaptive
functioning
and
behavioral
outcomes.
A
literature
search
identified
11
studies
suitable
for
inclusion
in
the
meta-analysis
on
adaptive
func-
tioning
(N
=
1,550),
24
studies
for
the
meta-analysis
on
internalizing
problems
(N
=
1,984),
21
studies
for
the
meta-analysis
on
externalizing
problems
(N
=
1,729)
and
25
studies
for
the
meta-analysis
on
total
behavior
problems
(N
=
2,523).
No
overall
improvement
or
dete-
rioration
was
found
for
adaptive
functioning.
However,
studies
with
a
timespan
longer
than
one
year
and
studies
with
larger
sample
sizes
showed
development
toward
more
negative
adaptive
functioning
than
studies
with
shorter
timespans
or
smaller
samples.
No
overall
increases
or
decreases
in
internalizing,
externalizing
or
total
behavior
problems
were
found.
Based
on
these
results,
it
is
concluded
that
foster
care
does
not
negatively
or
positively
affect
foster
children’s
developmental
trajectories.
Given
that
many
children
enter
foster
care
with
problems,
this
is
a
worrying
situation.
Further
longitudinal
research
to
find
the
factors
necessary
for
improving
foster
children’s
developmental
chances
is
recommended.
Furthermore,
routine
screening
and
targeted
foster-care
interventions
are
adviseable
to
ensure
that
all
children,
who
cannot
be
raised
by
their
own
parents,
receive
the
support
conducive
to
their
positive
development.
©
2015
Elsevier
Ltd.
All
rights
reserved.
Introduction
If,
for
any
reason,
children
cannot
be
raised
by
their
own
parents,
a
child
can
be
placed
in
a
foster
family.
Foster
care
has
been
suggested
as
a
better
alternative
than
residential
and
group
settings,
because
it
most
resembles
the
natural
home
environment
of
the
child
(Roy,
Rutter,
&
Pickles,
2000;
Webster,
Barth,
&
Needell,
1999;
Wilson
&
Conroy,
1999).
In
contrast
with
institutional
or
group
care,
which
is
characterized
by
a
great
discontinuity
of
caregivers,
foster
care
contains
the
stability
and
continuity
of
care
through
which
foster
children
can
build
a
close
relationship
with
their
foster
parents
(Tizard
&
Hodges,
1978).
Though
foster
care
may
be
considered
the
preferred
out-of-home
placement
option,
much
remains
unclear
about
the
development
of
children
in
foster
care
(Lawrence,
Carlson,
&
Egeland,
2006;
McWey,
Cui,
&
Pazdera,
2010;
Simmel,
Barth,
&
Brooks,
2007).
Several
studies
on
foster
children
point
at
severe
psychological
problems
at
the
start
of
placement
in
foster
Corresponding
author.
http://dx.doi.org/10.1016/j.chiabu.2015.02.003
0145-2134/©
2015
Elsevier
Ltd.
All
rights
reserved.
122
A.
Goemans
et
al.
/
Child
Abuse
&
Neglect
42
(2015)
121–134
families
(Clausen,
Landsverk,
Ganger,
Chadwick,
&
Litrownik,
1998;
Hochstadt,
Jaudes,
Zimo,
&
Schachter,
1987;
James,
2004;
Simms,
Dubowitz,
&
Szilagyi,
2000;
Zorc
et
al.,
2013)
and
at
high
rates
of
problems
emerging
in
foster
care
(Berkoff,
Leslie,
&
Stahmer,
2006;
Lloyd
&
Barth,
2011;
Minnis,
Everett,
Pelosi,
Dunn,
&
Knapp,
2006),
but
these
studies
provide
a
snapshot
and
do
not
give
insight
in
the
developmental
trajectories
of
foster
children.
To
understand
such
trajectories,
longitudinal
studies
are
needed
(Heath,
Colton,
&
Aldgate,
1994;
McWey
et
al.,
2010;
Simmel
et
al.,
2007;
Taussig,
2002).
Since
the
70s
and
80s
of
the
last
century,
researchers
conducted
longitudinal
studies
to
analyze
the
development
of
children
in
foster
care
(Fanshel
&
Shinn,
1978;
Frank,
1980).
Numerous
longitudinal
studies
have
been
performed
across
various
countries,
but
results
have
not
been
conclusive.
Some
studies
suggest
that
foster
care
placements
may
improve
children’s
functioning
(Ahmad
et
al.,
2005;
Barber
&
Delfabbro,
2005;
Fernandez,
2009;
White,
1997),
but
other
studies
reported
no
improvement
(Leathers,
Spielfogel,
McMeel,
&
Atkins,
2011;
Perkins,
2008)
or
even
deterioration
(Fanshel
&
Shinn,
1978;
Frank,
1980;
Lawrence
et
al.,
2006).
Because
large
numbers
of
children
over
the
world
experience
foster
care
each
year
(Stahmer
et
al.,
2009),
it
is
important
for
these
and
future
foster
children
to
obtain
a
clearer
picture
of
the
impact
of
foster
care
on
child
development.
The
challenge
of
inconclusive
findings
on
the
development
of
foster
children
may
be
resolved
in
a
meta-analysis.
Using
meta-analysis
an
overall
effect
size
of
studies
as
well
as
the
variance
of
effect
sizes
across
studies
can
be
analyzed
(Borenstein,
Hedges,
Higgins,
&
Rothstein,
2009).
In
the
current
study,
we
performed
a
series
of
four
meta-analyses
to
examine
the
longitudinal
developmental
outcomes
of
children
in
foster
care.
The
focus
is
on
adaptive
functioning
and
behavioral
outcomes
(internalizing,
externalizing
and
total
behavior
problems),
which
have
often
been
used
as
outcome
measures
in
longitudinal
studies
on
foster
care
(Barber
&
Delfabbro,
2005;
Fanshel
&
Shinn,
1978;
Minty,
1999;
Stahmer
et
al.,
2009).
We
expect
to
find
a
positive
development
of
foster
children
because
foster
care,
which
is
characterized
by
a
home-like
setting
and
continuity
of
care
that
cannot
be
offered
by
other
kinds
of
out-of-home-placements,
is
considered
as
second
best
in
absolute
sense
when
children
cannot
stay
with
their
parents
(Burns
et
al.,
2004).
Methodological
differences
in
design,
as
reflected
in
the
moderators,
are
expected
to
play
a
role
in
the
varying
outcomes
of
individual
studies.
Studies
on
the
development
of
foster
children
varied
in
study
length
(Frank,
1980;
White,
1997),
involved
small
samples
and
reported
considerable
attrition
across
waves
(Farmer,
Mustillo,
Burns,
&
Holden,
2008)
resulting
in
limited
generalizability
(Farmer
et
al.,
2008;
Maluccio
&
Fein,
1985).
These
methodological
differences
might
be
reflected
in
publication
type
(e.g.,
peer-reviewed
journals
are
less
likely
to
publish
studies
with
small
sample
sizes
and
insignificant
results
compared
to
book
chapters,
reports,
and
dissertations).
Furthermore,
the
age
of
foster
children
is
known
to
be
related
to
behavior
problems
(McWey
et
al.,
2010).
Moderator
analyses
will
therefore
be
performed
on
five
variables:
study
length,
sample
size,
attrition,
publication
type,
and
mean
age.
Method
Search
Strategy
Four
online
databases
were
used
to
systematically
search
for
longitudinal
studies
on
the
development
of
foster
children
published
until
2013.
ERIC,
Medline,
PsycINFO
and
ProQuest
Dissertations
&
Theses
were
searched
with
the
search
terms
foster
children
or
foster
care
combined
with
longitudinal,
repeated
measures,
or
pretest
posttest.
Second,
the
reference
lists
of
the
retrieved
articles,
dissertations,
books
and
reports
were
scanned
for
relevant
studies.
Third,
the
Doctoral
Dissertation
by
Van
Oijen
(2010)
and
reviews
by
Minty
(1999)
and
Maluccio
and
Fein
(1985)
were
scanned
for
further
potential
articles
for
inclusion
in
the
meta-analysis.
This
search
resulted
in
130
studies
which
included
articles,
reports,
dissertations
and
book
chapters.
A
flow
diagram
of
our
search
is
presented
in
Fig.
1.
Inclusion
and
Exclusion
Criteria
We
searched
for
studies
on
the
behavioral
and
adaptive
development
of
children
in
regular
foster
care.
Because
multiple
studies
include
several
outcomes
(Stahmer
et
al.,
2009;
Van
Oijen,
2010)
and
respondents
may
only
be
included
in
a
meta-
analysis
once,
separate
meta-analyses
will
be
conducted
for
adaptive
functioning,
internalizing
problems,
externalizing
problems,
and
total
problem
behavior.
Adaptive
functioning
is
generally
defined
as
meeting
age
and
culturally
appropriate
standards
of
personal
independence
and
social
functioning
(Horwitz,
Balestracci,
&
Simms,
2001).
Internalizing
behavioral
problems
are
considered
problems
that
primarily
affect
a
person
him
or
herself
and
are
characterized
by
emotional
symptoms
such
as
anxiety,
depression,
withdrawal
and
somatic
complaints.
Externalizing
behavior
problems
primarily
affect
a
person’s
social
environment
and
refer
to
delinquent
or
aggressive
behaviors
(Achenbach,
1991;
Horwitz
et
al.,
2001).
Studies
were
excluded
if
they
did
not
provide
longitudinal
data
on
these
outcome
measures.
Only
studies
that
reported
data
allowing
to
compute
an
effect
size
between
two
moments
of
measurement
were
included
in
the
meta-analysis.
This
means
that
a
study
had
to
provide
codeable
data
for
the
same
group
of
children
on
at
least
two
points
in
time,
though
between
the
two
time
points
there
were
sometimes
small
variations
in
sample
size
because
of
attrition
or
non-response.
The
length
of
time
that
elapsed
between
these
two
points
in
time
was
not
used
as
an
exclusion
criterium.
For
longitudinal
quasi-experimental
studies
with
an
intervention
and
control
group,
the
control
group
was
included
in
the
meta-analysis.
We
chose
to
include
control
groups
of
quasi-experimental
studies
because
they
provide
longitudinal
data
comparable
to
those
of
other
children
in
foster
care;
intervention
groups
were
excluded
because
the
intervention
may
have
changed
the
children’s
behavior
in
a
manner
that
is
atypical
for
foster
children
receiving
‘care-as-usual’.
For
the
same
reason,
studies
on
therapeutic
foster
care
or
foster
care
A.
Goemans
et
al.
/
Child
Abuse
&
Neglect
42
(2015)
121–134
123
Records identified through database
sea
rching
(n = 1020)
Additional records identified
throu
gh other
sources
(n = 28)
Records after duplicates removed
(n = 731)
Recor
ds screened
(n = 703)
Recor
ds excluded
(n = 573
)
Full-text arti
cles assessed
for
eligibili
ty
(n = 130)
Full-text articles excluded,
with reason
s
(n = 101)
Studies includ
ed in
quanti
tative
synthesis
(meta-an
alysis)
(n = 29)
Fig.
1.
Flow
diagram
of
all
stages
of
the
literature
search.
interventions
were
excluded.
Foster
children
in
all
age
groups,
from
infancy
to
adolescence,
were
included.
Excluded
were
studies
on
foster
children
older
than
18
years
at
the
start
of
the
study.
Furthermore,
some
studies
included
foster
children
as
well
as
other
kinds
of
out-of-home
placed
children
(e.g.,
adoptees,
institutionalized
children).
These
studies
were
included
only
when
they
reported
on
the
development
of
foster
children
separately,
and
for
this
information
only.
Some
studies,
clearly
longitudinal,
could
not
be
included
because
statistics
relevant
for
our
meta-analysis
were
not
reported
(Aarons
et
al.,
2010;
Vanderfaeillie,
Van
Holen,
Vanschoonlandt,
Robberechts,
&
Stroobants,
2013)
or
foster
children
and
children
in
other
kinds
of
out-of-home
placements
or
reunified
children
were
combined
(Horwitz
et
al.,
2001;
Raslaviciene
&
Zaborskis,
2002;
Slinning,
2004;
Stahmer
et
al.,
2009).
Authors
of
the
articles
reporting
on
these
studies
were
contacted
to
request
additional
information.
This
resulted
in
the
inclusion
of
four
(Bulat,
2010;
Minnis
et
al.,
2006;
Stahmer
et
al.,
2009;
Vanderfaeillie
et
al.,
2013)
additional
studies.
The
language
in
which
studies
were
written
was
not
used
as
an
exclusion
criterion,
thus
the
current
meta-analysis
is
not
limited
to
English
language
publications.
Four
Dutch
studies
were
included
in
the
analyses
(Bastiaensen,
2001;
Damen
&
Pijnenburg,
2005;
Damen
&
Veerman,
2005;
Van
Oijen,
2010).
One
article
in
German
was
excluded
because
the
age
range
of
the
participants
exceeded
18
years
at
the
start
of
the
study
(Linderkamp,
Schramm,
&
Michau,
2009).
Furthermore,
one
Lithuanian
(Raslaviciene
&
Zaborskis,
2002)
and
one
French
(Dumaret
&
Duyme,
1982)
article
were
excluded
because
they
both
focused
on
other
kinds
of
out-of-home
placements.
The
11
studies
summarized
in
Table
1
met
the
inclusion
criteria
for
the
meta-analysis
on
adaptive
functioning
of
foster
children
and
were
used
in
the
meta-analysis.
Table
2
gives
an
overview
of
the
29
studies
that
met
the
inclusion
criteria
for
the
meta-analyses
on
behavioral
functioning
(internalizing,
externalizing
or
total
behavioral
problems)
of
foster
children.
124
A.
Goemans
et
al.
/
Child
Abuse
&
Neglect
42
(2015)
121–134
Table
1
Studies
included
in
the
meta-analysis
on
adaptive
functioning.
Study
(year
of
publication) Country
N
(Attrition) Study
interval
Age
range
at
T1
(mean)
%
Female
Measure
(informant)
Ahmad
et
al.
(2005) aIraq
89
(5.30%)
24
months
6–18
years
(11.1)
50.00
CBCL
Total
Competence
(FP)
Barber
and
Delfabbro
(2005) aAustralia
109
(53.62%)
24
months
4–17
years
(10.8)
48.51
Social
adjustment
(prof.)
Bogart
(1988) b,cUSA
20
(n/a) 3–6
months 3–16
years
(9.8) 55.00 VABS
(FP),
CBCL
Total
Competence
(FP
&
prof.)
Fanshel
and
Shinn
(1978) aUSA
205
(63.39%)
5
years
0–12
years
(n/a)
49.36
CBC
(prof.)
Fernandez
(2008) aAustralia
39
(9.30%) 24
months 4–15
years
(8.8) 50.85 Adaptive
Functioning
(T)
Jacobsen
et
al.
(2013) aNorway
56
(6.67%)
12
months
22–25
months
(23.3months)
37.50
ITSEA
Competence
Domain
(FP)
Matthews
(1997) bUSA
48
(52.45%) 6
months
0.7–17.9
years
(n/a)
47.50
VABS
(FP)
McAuley
and
Trew
(2000) aNorthern
Ireland 12–15 (16.67–25.00%) 8
months 4–11
years
(8.4) 36.84 CBCL
Total
Competence
(FP)
Perkins
(2008) bCanada
201
(45.23%)
12
months
10–17
years
(12.9)
50.00
Pro
Social
Behavior
(FC)
Stahmer
et
al.
(2009) aUSA
758
(n/a) 21.3
months 0–15
years
(4.8) 52.83 VABS
(FP)
White
(1997) b,dUSA
10
(0.00%)
5
weeks
7–9
years
(8.5)
60.00
CBCL
Adaptive
Behavior
(FP)
FP
=
foster
parent,
prof.
=
professional,
T
=
teacher,
FC
=
foster
child.
aStudies
published
in
peer-reviewed
journals.
bStudies
published
in
non
peer-reviewed
journals.
cStudy
length
differs
for
type
of
informant
(foster
parents:
3
months,
social
workers:
6
months).
dExperimental
design.
A.
Goemans
et
al.
/
Child
Abuse
&
Neglect
42
(2015)
121–134
125
Table
2
Studies
included
in
the
meta-analyses
on
internalizing
(Int),
externalizing
(Ext)
and/or
total
behavior
problems
(Total).
Study
(year
of
publication)
Country
N
(Attrition)
Study
interval
Age
range
at
T1
(Mean)
%
Female
Measure
(informant)
Meta-analyses
Ahmad
et
al.
(2005) aIraq
89
(5.30%)
24
months
6–18
years
(11.1)
50.00
CBCL
(FP)
Int,
Ext,
Total
Barber
and
Delfabbro
(2005) aAustralia
109
(53.62%)
24
months
4–17
years
(10.8)
48.51
CBC
(prof.)
Int,
Ext,
Total
Bastiaensen
(2001) bNetherlands
53
(49.52%)
24
months
8–13
years
(10.4)
47.10
CBCL
(FP)
Int,
Ext,
Total
Bogart
(1988) b,cUSA
20
(n/a)
3–6
months
3–16
years
(9.8)
55.00
CBCL
(FP)
Total
Bulat
(2010) aCroatia
60
(46.43%)
5
years
10–17
years
(13.3)
50.00
CBCL
Anxiety
&
Depression
(FP),
YSR,
CDI
(FC)
Int
Damen
and
Pijnenburg
(2005) bNetherlands
51
(n/a)
6
months
0–18
years
(10)
50.53
CBCL
(FP)
Int,
Ext,
Total
Damen
and
Veerman
(2005) bNetherlands
41
(36.92%)
6
months
5–13
years
(9.5)
54.55
CBCL
(FP)
Int,
Ext,
Total
Fanshel
and
Shinn
(1978) bUSA
205
(63.39%)
5
years
0–12
years
(n/a)
49.36
CBC
(prof.)
Int,
Ext,
Total
Fernandez
(2008) aAustralia
51
(13.56%) 2
years
4–15
years
(8.8)
50.85
TRF
(T)
Int
Frank
(1980) a,dUSA
50
(0.00%)
5
years
6–12
years
(n/a)
n/a
Psychosocial
Problems
(prof.)
Total
Gonzalez
(1999) bUSA
15
(n/a)
12–24
months
2–11
years
(n/a)
53.33
CBCL
(FP)
Int,
Ext,
Total
Haight
et
al.
(2010) aUSA
7
(30.00%)
7
months
7–14.5
years
(9.6)
n/a
CBCL
(FP)
Int,
Ext,
Total
Jacobsen
et
al.
(2013) aNorway
56
(6.67%)
12
months
22–25
months
(23.3months)
37.50
ITSEA
(FP)
Int,
Ext,
Total
Lawrence
et
al.
(2006) a,eUSA
15
(n/a)
Pre-
and
postplacement
0–9
years
(n/a)
40.00
TRF
(T)
Int,
Ext,
Total
Leathers
et
al.
(2011) aUSA
6
(53.85%) 12
months
4–12
years
(8.6)
28.00
CBCL
(FP)
Int,
Ext,
Total
Leon,
Ragsdale,
Miller,
and
Spacarelli
(2008) aUSA
142
(n/a)
6
months–3.2
years
10.4–17.9
years
(13.2)
27.00
Negative
Affect
(FC)
Int
Linares,
Li,
Shrout,
Brody,
and
Pettit
(2007) aUSA
156
(n/a)
12
months
3–14
years
(8.4)
42.31
Loneliness,
CDI
(FC),
ECBI
(P)
Int,
Ext,
Total
Love
et
al.
(2008) a,f USA
20–22
(4.55%/13.04%)
6
months
6–17
years
(n/a)
60.87
BAI
(FC),
CBCL
(FP),CDI
(FC)
Int,
Ext,
Total
Matthews
(1997) bUSA
27
(73.27%) 6
months
1.8–6.7
years
(n/a)
47.50
CBCL
(FP)
Int,
Ext,
Total
McAuley
and
Trew
(2000) a,gNorthern
Ireland
12–15
(16.67–25.00%)
8
months
4–11
years
(8.4)
36.84
TRF
(T)
CBCL
(FP)
Int,
Ext,
Total
McWey
et
al.
(2010) aUSA
106
(85.42%)
3
years
13–16
years
(14)
52.00
CBCL
(FP)
Int,
Ext
Minnis
et
al.
(2006) aScotland
88
(16.98%)
9
months
5–16
years
(11.6)
44.34
SDQ
(FP)
Total
Newton
et
al.
(2000) aUSA
415
(10.75%)
12
months
2–17
years
(6.6)
53.50
CBCL
(FP)
Int,
Ext,
Total
Perkins
(2008) bCanada
201
(45.23%) 12
months 10–17
years
(12.9) 50.00
Emotional
disorder,
conduct
disorder,
indirect
aggression
(FC)
Int,
Ext,
Total
Rushton
et
al.
(1995) a,hUK
12/15
(33.33%/16.67%)
5–8
years
5–9
years
(6.8)
0.00
Parental
interview
(FP)
Int,
Ext,
Total
Stahmer
et
al.
(2009) aUSA
752
(n/a)
21.3
months
0–15
years
(4.8)
52.83
CBCL
(FP)
Total
Van
Oijen
(2010) b,iNetherlands
78/59
(15.22%/15.38%) 21–24
months n/a
(14.4)
56.50
CBCL
(FP),
YSR
(FC)
Int,
Ext,
Total
Vanderfaeillie
et
al.
(2013) aBelgium
49
(36.36%)
24
months
6–12
years
(9.3)
63.27
CBCL
(FP)
Int,
Ext,
Total
White
(1997) b,jUSA
10
(0.00%)
5
weeks
7–9
years
(8.5)
60.00
CBCL
(FP)
Total
FP
=
foster
parent,
prof
=
professional,
T
=
teacher,
FC
=
foster
child.
aStudies
published
in
peer-reviewed
journals.
bStudies
published
in
non
peer-reviewed
journals.
cStudy
length
differs
for
type
of
informant
(foster
parents:
3
months,
social
workers:
6
months).
dPercentage
female
is
not
reported.
However,
according
to
the
authors,
the
sample
is
a
balanced
representation
according
to
sex.
eLength
of
placement
differs
from
1
to
45
months
due
to
the
design
of
this
study:
pre-post
placement
measurement.
fSample
size
and
attrition
differs
for
type
of
instrument:
BAI
&
CDI,
N
=
22
(4.55%),
CBCL,
N
=
20
(13.04%).
gAttrition
rate
differs
for
type
of
informant
(mother:
21.05%,
father:
25.00%,
teacher:
16.67%).
hSample
size,
attrition,
and
study
length
differs
for
type
of
instrument:
Rutter
A
scales,
N
=
12
(33.33%),
8
years.
Parental
interview,
N
=
15
(16.67%),
5
years.
iSample
size
and
attrition
rate
differs
for
type
of
informant/instrument
(foster
parents/CBCL:
N
=
78,
attrition
rate
=
15.22%,
foster
child/YSR:
N
=
59,
attrition
rate
=
15.38%).
jExperimental
design.
126
A.
Goemans
et
al.
/
Child
Abuse
&
Neglect
42
(2015)
121–134
Coding
Decisions
and
Extraction
of
Effect
Sizes
We
always
used
two
time-points
based
on
the
same
sample
of
children
to
compute
Hedges
g.
Hedges
g
is
an
effect
size
measure
like
Cohen’s
d
but
is
computed
slightly
different,
incorporating
an
adjustment
which
removes
the
bias
of
Cohen’s
d.
Hedges
g
is
defined
as
the
difference
between
the
two
means,
divided
by
the
pooled
standard
deviation
(Borenstein
et
al.,
2009).
If
an
article
provided
more
than
two
time
points,
we
choose
to
code
the
two
time
points
that
were
farthest
apart
(the
first
and
the
last
moment
of
measurement).
If
articles
included
multiple
independent
samples
(e.g.,
foster
chil-
dren
from
different
foster
care
institutions;
boys
and
girls)
these
were
entered
in
the
meta-analyses
separately.
If
articles
included
multiple
dependent
samples
(e.g.,
multiple
informants
for
the
same
study
population)
the
findings
were
aver-
aged
in
the
meta-analyses
(Borenstein
et
al.,
2009),
also
when
study
length
differed
for
type
of
informant
(Bogart,
1988;
Rushton,
Treseder,
&
Quinton,
1995).
Furthermore,
several
studies
distinguished
several
aspects
of
the
general
domain
of
adaptive
or
behavioral
functioning,
but
did
not
give
an
overall
score
(Fanshel
&
Shinn,
1978;
Perkins,
2008).
In
these
stud-
ies,
the
statistics
for
different
aspects
of
adaptive
or
behavioral
functioning
were
combined
into
one
general
estimator
of
that
domain
(Goodman,
1997).
Two
studies
(McWey
et
al.,
2010;
Stahmer
et
al.,
2009)
used
the
same
data
for
fos-
ter
children.
Because
samples
included
in
meta-analysis
must
be
independent,
both
articles
could
not
be
included
in
the
same
meta-analysis.
One
of
these
studies
is
included
in
the
meta-analysis
on
adapative
functioning
and
total
behavioral
problems,
but
did
not
report
internalizing
and
externalizing
behavior
problems
separately
(Stahmer
et
al.,
2009).
There-
fore,
the
other
study
is
included
in
the
meta-analyses
with
respect
to
each
of
these
(McWey
et
al.,
2010).
If
the
same
author(s)
published
more
than
one
article
on
the
same
study
sample,
we
chose
to
include
the
study
with
the
larger
sample
size.
Information
on
the
test–retest
correlations
is
often
missing
from
articles
presenting
longitudinal
data
(Morris
&
DeShon,
2002).
In
many
meta-analyses
that
analyze
longitudinal
data
the
problem
of
missing
correlations
is
resolved
by
treating
the
means
as
independent
(Jones,
Riley,
Williamson,
&
Whitehead,
2009).
This
approach
is
likely
to
result
in
a
biased
meta-
analysis
(Mavridis
&
Salanti,
2012;
Morris
&
DeShon,
2002;
Riley,
2009),
and
thus
we
chose
to
include
test–retest
correlations
in
our
calculation
of
effect
sizes
to
account
for
the
dependent
nature
of
the
data.
To
solve
the
problem
of
unreported
test–retest
correlations,
the
use
of
an
aggregate
test–retest
correlation
coefficient,
wherein
available
correlation
coefficients
are
aggre-
gated,
is
advised
as
an
option
to
account
for
the
dependent
nature
of
the
data
in
a
meta-analysis
based
on
longitudinal
data
(Morris
&
DeShon,
2002).
We
coded
test–retest
correlation
coefficients,
or
obtained
test–retest
correlation
coefficients
from
paired
sample
t-tests
(Cooper,
Hedges,
&
Valentine,
2009).
Two
of
the
authors
independently
coded
the
means
and
standard
deviations
for
both
time
points
as
well
as
the
sample
size
and
the
moderators.
The
coded
moderators
were
study
length,
sample
size,
publication
type,
attrition,
and
mean
age.
Differences
between
authors
were
resolved
by
discussion.
Prior
to
discussion,
the
authors
coded
identically
90
percent
of
the
time.
Analyses
Data
were
analyzed
using
the
program
Comprehensive
Meta-Analysis
(Borenstein,
Hedges,
Higgins,
&
Rothstein,
2005).
Raw
mean
difference,
standard
deviation
of
both
time
points,
aggregated
correlation
and
sample
size
were
used
to
compute
effect
sizes.
For
two
studies
(Fernandez,
2008;
Love,
Koob,
&
Hill,
2008)
raw
mean
difference
and
paired
t
were
used
to
compute
effect
sizes.
Four
meta-analyses
were
conducted
to
examine
the
developmental
outcomes
of
foster
children
with
respect
to
four
domains:
adaptive
functioning,
externalizing
behavior
problems,
internalizing
behavior
problems
and
total
problems.
Data
were
analyzed
using
a
random
effects
model,
which
does
not
assume
a
common
underlying
effect
size
for
all
included
studies,
and
is
commonly
more
appropriate
for
meta-analyses
based
on
a
literature
search
than
a
fixed
effects
model
(Borenstein
et
al.,
2009).
The
Q
test
is
used
to
test
whether
studies
are
homogeneous.
A
significant
Q
test
suggests
true
heterogeneity
between
included
effect
sizes.
The
I2was
used
to
quantify
the
heterogeneity
between
the
effect
sizes
of
included
studies;
the
I2can
be
interpreted
as
the
percentage
of
total
variability
in
a
set
of
effect
sizes
due
to
true
heterogeneity
(Huedo-Medina,
Sánchez-Meca,
Marin-Martinez,
&
Botella,
2006).
If
I2is
large,
then
it
would
make
sense
to
speculate
about
reasons
for
the
variance,
and
possibly
to
apply
techniques
such
as
subgroup
analysis
to
try
to
explain
it.
Moderator
analyses
will
be
performed
on
five
variables:
study
length,
sample
size,
attrition,
publication
type,
and
mean
age.
The
Q
statistic
was
used
to
test
for
the
significance
of
moderators.
The
jackknife
procedure
is
used
to
identify
studies
with
large
influence
on
the
overall
effect
size
estimate.
This
procedure
gives
insight
whether
the
overall
effect
size
is
biased
by
the
influence
of
any
one
study
(Borenstein
et
al.,
2009).
Furthermore,
because
non-significant
results
may
be
missing
in
the
studies
sampled,
due
to
publication
bias,
the
effect
sizes
computed
in
the
meta-analysis
may
be
overestimated.
To
assess
the
risk
of
such
publication
bias,
we
used
Duvall
and
Tweedie’s
trim-and-
fill
procedure
the
Kendall
method.
Duvall
and
Tweedie’s
trim-and-fill
procedure
is
an
iterative
procedure
that
imputes
effect
sizes
until
the
error
distribution
closely
approximates
normality.
This
procedure
provides
a
more
unbiased
estimate
of
the
effect
size
than
the
observed
estimate.
The
association
between
the
standardized
effect
sizes
and
the
variance
of
these
effect
sizes
was
calculated
using
the
Kendall
method.
A
significant
correlation
indicates
that
small
studies
with
non-significant
results
tend
not
to
be
published,
which
suggests
that
publication
bias
exists.
A
non-significant
Kendall
coefficient
suggests
the
absence
of
such
publication
bias.
A.
Goemans
et
al.
/
Child
Abuse
&
Neglect
42
(2015)
121–134
127
Study name S tati sti cs for e ach study
Hedges 's g a nd 95% CI
He dges 's
Lo we r
Upp
er
gli
mit
limit
Ahmad et al. (2005
)
-1,18
-1,50
-0,86
Barber & Delfabbro (200
5)
-0,43
-0,66
-0,20
53,052,0-50,0)8
8
91( t
r
ag
oB
Fanshel & Shinn (197
8)
-0,07
-0,17
0,03
Fernandez (2008)
1,08
0,61
1,55
Jac
obsen et al. (20
13)
0,18
-0,13
0,49
Matt
hews (1997)
0,06
-0,27
0,40
McAuley & Trew (200
0)
0,15
-0,21
0,51
Perkins (2008)
-0,06
-0,23
0,10
Stahmer et al. (2009
; kin)
-0,46
-0,60
-0,33
Stahmer et al. (2009
; non
-kin)
-0,55
-0,67
-0,44
2
9,
11
1,
020
,
1)
7
9
9
1(
eti
h
W
-0,10
-0,32
0,12
-2,00
-1,00
0,00
1,00 2,00
Fig.
2.
Forest
plot
for
the
meta-analysis
on
adaptive
functioning.
Table
3
Results
of
the
meta-analysis
on
adaptive
functioning,
internalizing
behavior
problems,
externalizing
behavior
problems
and
total
behavior
problems.
k
Samples
N
g
CI
I2Q
Adaptive
functioning 11
12
1,550
.10
.32,
.12
92.35
143.86
Internalizing
behavior
problems
24
34
1,984
.10
.27,
.07
91.47
386.78
Externalizing
behavior
problems
21
29
1,729
.04
.24,
.15
91.28
321.16
Total
behavior
problems
25
35
2,523
.10
.28,
.07
94.36
602.50
Results
Meta-analysis
on
Adaptive
Functioning
Eleven
studies
eligible
for
inclusion
yielded
twelve
effect
sizes
(Table
1).
These
11
studies
reported
on
the
development
of
a
total
of
1,550
foster
children,
ranging
in
age
from
0
to
18
years.
Analyses
revealed
no
overall
significant
difference
between
the
effect
sizes
of
both
measurement
points
(g
=
.10,
p
=
.38,
N
=
1,550).
This
means
that
this
meta-analysis
showed
no
improvement
or
deterioration
in
the
adaptive
functioning
of
foster
children
during
their
stay
in
foster
care.
The
studies
included
in
the
meta-analysis
were
highly
heterogeneous
(Q(10)
=
143.86,
p
<
0.001),
and
a
fairly
high
proportion
of
the
observed
variance
reflected
real
differences
in
effect
size
(I2=
92.35),
meaning
that
studies
included
in
this
meta-analysis
tended
to
provide
different
effect
sizes
(Higgins,
Thompson,
Deeks,
&
Altman,
2003).
The
forest
plot
provided
in
Fig.
2
gives
a
graphic
depiction
of
the
effect
sizes
of
the
included
studies.
In
this
forest
plot,
a
positive
effect
size
corresponds
with
an
improvement
in
adaptive
functioning.
The
Duvall
and
Tweedie
trim-and-fill
procedure
showed
that
four
studies
to
the
left
of
the
mean
needed
to
be
imputed
to
shift
the
observed
point
estimate
from
.10
(CI:
.32,
.12)
to
.32
(CI:
.54,
.10),
which
suggests
that
four
studies
with
significant
deterioration
of
adaptive
functioning
would
result
in
a
significant
overall
effect.
The
jackknife
procedure
showed
that
the
overall
effect
remained
the
same
when
one
study
at
a
time
was
removed
from
the
meta-analysis.
The
Kendall’s
was
.26
(z
=
1.17,
p
=
.12),
which
suggests
the
absence
of
publication
bias.
Results
are
displayed
in
Table
3.
Five
moderator
analyses
were
performed
to
compare
studies
on
the
methodological
characteristics
study
length,
sample
size,
attrition,
publication
type,
and
mean
age.
Studies
with
a
timespan
less
than
one
year
(g
=
.08,
p
=
.33,
k
=
6,
N
=
350)
and
studies
with
a
timespan
of
one
year
or
longer
(g
=
.31,
p
=
.06,
k
=
5,
N
=
1,200)
yielded
a
significant
difference,
Q(1)
=
4.62,
p
<
.05.
This
means
that
studies
following
children’s
development
during
a
longer
timespan
showed
more
negative
adap-
tive
functioning
development
than
studies
with
a
shorter
timespan.
Furthermore,
studies
with
large
sample
sizes
(N
80)
(g
=
.44,
p
<
.01,
k
=
5,
N
=
1,362)
and
studies
with
small
sample
sizes
(N
<
80)
(g
=
.33,
p
<
.05,
k
=
6,
N
=
188)
also
<