Content uploaded by Tonino Esposito
Author content
All content in this area was uploaded by Tonino Esposito on Apr 06, 2016
Content may be subject to copyright.
88 Copyright © 2015 International Journal of Child and Adolescent Resilience
International Journal of Child and Youth Resilience
The Stability of Residential and Family Foster Care in
Quebec, Canada: A Propensity Weighted Analysis
Tonino Esposito1, Martin Chabot2, Ashleigh Delaye2,
and Nico Trocmé2
1 École de service social, Université de Montréal
2 School of Social Work, McGill University
Abstract:
Objectives: This province-wide analysis examined factors most associated with changing
out-of-home placements for 15,518 youth aged 10 to 17 at initial placement. This analysis
allows the stability of residential and family foster care to be more precisely examined.
Methods: This analysis draws clinical administrative data from all of Quebec’s child
protection agencies and the 2006 Canadian Census. Applying a method of quasi-
randomization using propensity weights that control for differences in the needs of
youth placed in residential and family foster care at initial placement, multivariate logistic
regression models were used to examine the risk of changing placements.
Results: At initial placement, youth manifesting severe behavioral problems are 431%
more likely to be admitted to residential care and 113% more likely if they committed
a crime prior to initial placement. While the analysis was weighted using propensity
estimates, those placed in residential care are 15% more likely to experience a total of
one placement change, 72% more likely to experience a total of two placement changes,
and 87% more likely to experience at least three placement changes compared to
their counterparts in family foster care. In addition, the risk of disruption increases in
magnitude for those with multiple investigations, longer spells of out-of-home care, and
who manifest high risk behaviors including youth criminality. Combined, these factors
make these youth the most likely to experience placement disruption than any other
youth placed in out-of-home care.
Implications: Given its’ inherent instability, residential care should be used only when
other alternatives, such as family foster care or in home services, are not possible.
89
Volume 1, Number 1, 2013, pp. Volume 3, Number 1, 2015, pp.
Keywords:
Out-of-home placement, placement stability, clinical-administrative data, census data,
propensity analysis.
Acknowledgement:
Funding is acknowledged from the Canada Research Chair in Social Services for
Vulnerable Children; Insight Development Grant from the Social Sciences and Humanities
Research Council (430-2014-00299); and a Team grant funding from the Canadian
Institutes of Health Research in boys’ and men’s health (TE3 138302).
Introduction
Residential care is a measure of last resort and can provide the safety, structure and
stability needed when there is no family willing or able to care for the troubled youth. While
it is assumed that family foster care is a more stable placement, it remains unclear as to what
extent youth in residential care, compared to youth in family foster care, can successfully
experience the stability needed to support their developmental needs. The association
between initial placement type and later stability is difficult to understand primarily because
of issues that disqualify the use of randomised control trials between placement types. As
such, there is limited information controlling for the differences between youth entering
residential care compared to those entering family foster care, and the effectiveness of the
placement type in ensuring later stability (Barth 2002; Souverein, Van der Helm & Stams
2013; De Swart et al. 2011; 2012).
This paper builds on the longitudinal work of Esposito and colleagues (2014), which
reports that close to one third (29.8%) of youth aged 10 to 17 years at initial placement
in Quebec experience multiple placement disruptions. Esposito et al. (2014) reported
that the increased risk of placement changes for these youth was primarily explained by
a combination of multiple child protection investigations, behavioral problems, school
difficulties, residential care, youth criminality, and socioeconomic disadvantages (Esposito
et al. 2014). This paper builds on this analysis by examining the extent to which these factors
continue to impact the likelihood of placement disruption after applying a method of quasi-
randomization using propensity weights, which control for differences in the needs of youth
placed in residential and family foster care at initial placement.
Background
Connell and associates (2006) note some inconsistency in the scholarly literature
regarding factors leading to placement disruption due to variation in the types of out-
of-home care studied. In their longitudinal study attempting to clarify some of these
inconsistencies, Connell et al. (2006) examined all types of out-of-home care settings and
© Esposito, Chabot, Dalaye, and Trocmé
88-100
90 Copyright © 2015 International Journal of Child and Adolescent Resilience
factors leading to placement disruption. The results of this study were more or less consistent
with the literature suggesting that older age is the primary factor associated with disruption.
Findings from Connell et al. (2006) differed somewhat from other scholarship centred on the
unanticipated results that youth with identified behavioral difficulties are at no greater risk
for placement disruption than youth experiencing neglect. However, Connell et al. (2006)
later confirms in a post-hoc analysis that youth with behavioral problems have higher rates
of placement change. Connell et al (2006) further suggested that youth placed in residential
care are two-and-a-half to nine times - depending on degree of restrictiveness - more likely
to experience placement changes when compared to children in relative foster homes. Lastly,
the authors found that while one previous move had minimal effect on later stability, two or
more removals significantly decreased the odds of placement stability; a finding that is also
described elsewhere (Farmer et al., 2008, Park & Ryan 2009).
Leathers’ (2006) study focused specifically on family foster care and the associations
between a youth’s externalized behavioral problems while in care and later stability, finding
that behavioral problems was a strong predictor of future negative outcomes (residential
treatment, imprisonment and runaway status). Leathers, (2006) suggested behavioral
problems in youth leading to disruption and negative outcomes could be mitigated by good
foster home integration. Although relevant to the present study, Leathers’ study is limited in
that the stability of the sample is not compared to the stability of youth in residential care.
Focusing on factors leading to placement in family foster care to more restrictive
residential care, Farmer and associates (2008) found that a youth’s difficulties at the time
of placement influenced the restrictiveness of the placement. The study measured youth
behavior using the Child Behavior Checklist (CBCL), and reported that youth with the
highest CBCL scores were more likely to be placed in restrictive placement settings. Farmer et
al. (2008) reported that on average, youth placed in more restrictive settings also experienced
more placement disruption than those placed in less restrictive settings; findings in line
with an earlier published meta-analysis by Oosterman and associates (2007), who reported a
significant effect (r = .18) of placement in residential care with placement disruption. Factors
leading to placement were also explored in Park and Ryan’s (2009) study, which examined the
stability of youth in both residential and non-residential care settings. The finding was that
youth with a history of inpatient mental health treatments were more likely to experience
residential care as a first out-of-home placement, be older, be frequently moved, and run
away from out-of-home care. The study also indicated that there is an association between an
increasing length of time youth were in out-of-home care and decreasing odds of placement
stability.
In a more detailed examination of children placed in residential care, James and
associates (2006) report that older males with clinically significant behavior problems were
more likely to be placed in residential care settings. However, unlike Farmer et al. (2008)
and Park & Ryan’s (2009) study, James and associates found that youth who were placed in
restrictive settings at first placement had less placement disruption. James and associates
(2006) also report that multiple placements, irrespective of placement setting is a significant
predictor of further disruption.
91
Volume 1, Number 1, 2013, pp. Volume 3, Number 1, 2015, pp.
While these studies contribute to our understanding of factors associated to placement
disruption, they do not lend themselves methodologically to understanding the effects
of placement type on later stability. This is primarily because placement studies do not
methodologically address differences in severity of behavioral difficulties and other
associated vulnerabilities between those placed in residential and family foster care at initial
placement, which together results in selection bias and confounds the associations often
made on the effectiveness of placement type or primary factors leading to later disruption.
The objective of this paper, therefore, is to understand the effect of out-of-home placement
type on later stability and factors associated with disruption using propensity weighted
models that more rigorously adjust for selection bias at initial placement.
Method
This study uses a multivariate research design and propensity weights to control for
pre-placement differences between youth placed in residential care and youth placed in foster
care. The study draws data from two sources: (1) clinical administrative data from Quebec’s
child protection agencies; and, (2) 2006 Canadian Census data. The first data source consists
of anonymized longitudinal clinical administrative child protection data from all mandated
child protection regions across the province of Quebec. These data were drawn from a
common provincial information system used by every mandated child protection agency in
Quebec and contain data on approximately 450,000 children dating back to 1989 (Esposito
et al. 2015). All covariates used in this study – except for neighborhood socioeconomic
disadvantages – were constructed using these clinical administrative data. The second data
source is provincial data extracted from the 2006 Canadian Census public archive at McGill
University, used to create a neighborhood socioeconomic disadvantage composite index.
Out-of-home placement changes includes moves in: (1) a formal subsidized placement
in family-based care; and, (2) a formal subsidized placement in a structured group living
setting or a therapeutic residential treatment facility. Each model is divided into two discrete
groups: (1) youth who do not change placements; and, (2) youth who change placements a
total of one time, two times, or at least three times. However, categorizing placement change
sequences in distinct groups assumes that all the children in the cohort had completed their
placement spells. While this is the case for the vast majority of children, the full history of
placement changes is not accounted for some youth whose initial placement occurred at the
tail end of the data coverage period (September 2011). To ensure that these cases did not bias
our models, we reduced the cohort from 18,468 youth, 10-17 years of age, placed for the first
time between April 2002 and September 2011 to 15,518 youth of the same age who exited
their last placement and child protection services before June 2011.
Covariates
Covariates examined in this study include age at initial placement, gender, the reason
for investigation at initial placement, the number of investigations, the source of referral
at initial placement, request for youth criminal justice services, and the neighborhood
socioeconomic disadvantages. These covariates were used in several multivariate logistic
© Esposito, Chabot, Dalaye, and Trocmé
88-100
92 Copyright © 2015 International Journal of Child and Adolescent Resilience
regression models with propensity weights, in order to obtain the independent effect of each
covariate on the risk of experiencing an out-of-home placement change.
Age at initial placement is a nominal variable with youth aged 10 to 13 acting as a
reference group for 14 to 17 year olds. Gender is a nominal variable with female acting as
the reference group for male. Reason for investigation prior to initial placement includes the
following dichotomous constructs: (1) psychological and emotional abuse, which includes
rejection, denigration, exposure to intimate partner violence and exploitation; (2) physical,
material, and health neglect, which includes physical neglect, medical neglect and material
deprivation; (3) parents’ high risk lifestyle resulting in a failure to supervise or protect the
youth, including substance use, abandonment due to parental absence, refusal to assure care,
and risk of neglect; (4) school truancy and school neglect, which includes failure to attend
school or failure to ensure the youth attends school; (5) risk of, or actual physical abuse —
reference category; (6) risk of, or actual sexual abuse; and, (7) behavioral problems such as
harming behavior, violence towards self and others, youth substance abuse, running away
and destruction of property. Youth criminal justice service request is a nominal variable
measuring whether youth received a request for services under the Quebec Youth Criminal
Justice Act (LSPJA — Loi sur la justice pénale des adolescents) prior to the first placement
change. Number of investigations is a continuous variable calculated by examining the
number of times youth are investigated by child protection prior to the placement change or
case closure. Source of referral at initial placement includes the following nominal variables:
(1) community health and social services clinic (CLSC); (2) child protection agency—
includes investigations by the same agency resulting from new allegations; (3) extended
family and neighbors—reference category; (4) school staff; (5) police; (6) hospital staff; (7)
other professional institutions; and, (8) unknown. Initial placement type is a nominal variable
measured at initial placement with family foster care acting as the reference group for
residential or group care. Given that the clinical-administrative data used in this study does
not allow us to differentiate between residential placement settings or between relative and
non-relative care, residential care includes youth initially placed in either a structured group
living setting or a therapeutic residential facility and, family foster care refers to those initially
placed in a subsidized relative or non-relative family. Cumulative time in out-of-home care
represents the sum of all out-of-home care spells for the clinical population studied.
Poverty plays a particularly important role in the stress experienced by youth. The stress
of living in high poverty environments create additional psychological demands that aggravate
the challenges these troubled youth face, and affects their decision making abilities. As such,
it is particularly important to control for neighborhood socioeconomic vulnerabilities in
predicting the propensity to initial placement type. A neighborhood socioeconomic index
was created using six socioeconomic indicators for each census dissemination area in Quebec.
They are: (1) total population 15 years and over who are unemployed or not in the labor force;
(2) median income in 2005 for the population 15 years and over; (3) total persons in private
households living alone; (4) total population 15 years and over who were separated, divorced
or widowed; (5) family median income in 2005; and, (6) median household income in 2005.
A principal components analysis with varimax rotation was performed on all the transformed
and normalized census-based indicators in order to construct a single index representing the
93
Volume 1, Number 1, 2013, pp. Volume 3, Number 1, 2015, pp.
socioeconomic neighborhood disadvantages for each dissemination area. This composite index
was then merged with the child protection clinical-administrative data matched by children’s
postal codes at initial investigation. The index has a minimum score of −3.37 representing
the lowest socioeconomic risk and a maximum score of 3.51 representing the highest
socioeconomic risk.
Analytic model
The analysis is composed of several steps. First, descriptive analyses were performed
between all independent covariates and number of placement changes (see Table 1). Ordinary
least squares linear regression was conducted in order to determine the variance inflation
factor (VIF), which ensures that there is no linearity among independent covariates. If the
values of VIF exceed 5, they are regarded as indicating multi-collinearity (O’Brien, 2007).
Individual Factors
Youth in out-of-
home care
10-17
(N=15,518)
Youth in family
foster care
10-17
(N=6,401)
Youth in residential
care
10-17
(N=9,117)
Age at initial placements:
10-13
14-17
25.8%
74.2%
36.0%
64.0%
18.6%
81.4%
Gender:
Male
Female
51.3%
48.6%
41.0%
49.0%
58.7%
41.3%
Reason for initial placement:
Psychological & emotional abuse
Physical, material & health neglect
Parent high risk lifestyle
School truancy & neglect
Risk of or sexual abuse
Behavioural problems
Risk of or physical abuse
4.2%
1.6%
16.6%
3.1%
4.0%
58.6%
11.9%
7.3%
2.3%
27.2%
3.1%
6.4%
33.1%
20.7%
2.0%
1.0%
9.2%
3.2%
2.4%
76.4%
5.8%
Source of referral at initial placement:
CLSC
Youth protection agency
Police
Other professional institutions
School
Hospital staff
Unidentified
Family
9.4%
9.1%
23.8%
4.5%
16.3%
4.7%
2.1%
30.1%
10.6%
9.7%
21.4%
4.6%
18.9%
3.5%
2.9%
28.2%
8.6%
8.5%
25.4%
4.4%
14.5%
5.5%
1.5%
31.4%
Request for youth criminal justice services: 21.4% 11.0% 28.7%
Mean (S.D.) Mean (S.D.) Mean (S.D.)
Number of placement changes 2.08(2.74) 1.91(2.66) 2.19(2.79)
Cumulative time in out-home care 447(503) 504(574) 407(442)
Number of investigations 1.61(.95) 1.67(1.02) 1.56(.90)
Socioeconomic disadantage .22(93) .34(.90) .15(.95)
Table 1: Descriptive Factors
© Esposito, Chabot, Dalaye, and Trocmé
88-100
94 Copyright © 2015 International Journal of Child and Adolescent Resilience
The VIF estimates ranged from a low of approximately 1.026 to a high of 2.891, indicating no
issues of linearity between covariates. Next, a multivariate logistic regression model was used
to reduce conceptually related indicators representing possible differences between youth
in residential and family foster care (age at initial placement, gender, behavioral problems,
youth criminal justice service request prior to placement, number of investigations prior to
placement, and socioeconomic disadvantages) to a predicted propensity score of residential
placement (see Table 2). The propensity score is defined as the probability of placement in
residential care based on the measured covariates listed above. Based on the explanation
given by Thoemmes, F. (2012), the propensity score is specified as:
ê(x) = P(Z=1 | X)
where ê(x) is the notation for propensity score, P a probability, Z=1 is the placement
type with values 0 for family foster care and 1 for residential care, conditional on “|” the
covariates used to compute the propensity (X). Therefore the propensity score expresses
the probability that a youth is to be placed in residential care. Propensity weights were
then developed based on the propensity score. As suggested by Guo & Fraser (2010), the
propensity weight for youth in residential care is 1/ê(x) and the propensity weight for youth
in family foster care is [1/ (1-ê(x))], inflating the propensity weight to predict placement
changes for youth in family foster care. The inverse propensity weight is then normalized to
sum to the clinical population studied. The normalized propensity weight is equal to the ratio
of the size of the clinical population studied to the sum of the inverse propensity weights.
Multivariate logistic regressions with propensity weights were then used to examine the
risk of changing placements. The multivariate models reported in Table 3 present the odds
ratio adjusted using inverse propensity weights and Wald statistic which allows us to quickly
consider whether the null hypothesis that the true coefficients equals zero. The dataset was
constructed and analyzed using SPSS version 22 and statistical tests were conducted at 95%
level of confidence.
Results
The data revealed that of the 15,518 youth aged 10 to 17 years at initial placement,
58.8% (N = 9,117) were placed in residential care and 41% were placed family foster care (N =
6,401). Of the youth initially placed in family foster care, 21.7% are placed in residential care
at some point in their placement spell, while 7.7% of youth initially admitted to residential
care change to family foster care. While it is assumed in this study that any placement change
– except family reunification – is potentially harmful from a developmental perspective,
moving from residential care to family foster care was considered to be a potentially positive
move. As such, propensity weights and multivariate models excluding youth initially placed
in residential care who moved to family foster care (N = 698) were used to examine the
sensitivity to changes in predictors of placement changes – the significance and direction of
estimates collate fully with those reported in tables 2 & 3.
The average cumulative placement duration for youth in foster family care was 504
days, while the average cumulative placement duration for youth in residential care was 407
95
Volume 1, Number 1, 2013, pp. Volume 3, Number 1, 2015, pp.
days. A third of youth did not experience a placement change (33%), while 24% changed
placements a total of one time, 14% a total of two times, and 29% changed placements 3 or
more times. As shown in Table 1, the majority of youth were aged 14 to 17 years at initial
placement, with a higher proportion of that cohort placed in residential care at initial
placement. There is a relatively equal proportion of male and female youth placed out-of-
home; however female youth were more likely to be placed in family foster care), whereas
male youth were more likely to be placed in residential care. Close to two-thirds of youth
placed in out-of-home care were investigated for severe behavioral problems (58.6%),
followed by 16.6% investigated because of parents’ high risk lifestyle. Among youth placed in
family foster care, 33.1% were investigated for behavioral problems, 27.2% for parents’ high
risk lifestyle, and 20.7% investigated for physical abuse. For youth placed in residential care,
76.4% were investigated for severe behavioral problems, 9.2% for parents’ high risk lifestyle
and less than 6% for physical abuse; a notable difference compared to youth placed in family
foster care.
Overall, the highest proportions of placed youth were reported by a family member
(30.1%), followed by the police (23.8%), and then school staff (16.3%); figures that represent
an overall tendency that remains consistent for youth placed in family foster care and
residential care. However, there is a higher proportion of youth placed in residential care
reported by the police. Close to one in five youth placed out-of-home had a request for youth
criminal justice services prior to placement; a rate that drops for youth placed in family foster
care, but increases for youth placed in residential care. The average number of placement
changes for all youth, irrespective of placement setting is 2.08 (std. 2.74), an average that
decreases for youth in family foster care but increases to 2.19 (std. 2.79) for youth in
residential care. The average count of child protection investigations is 1.61 (std. 95) per
youth, an average which increases slightly for youth placed in family foster care and decreases
slightly for youth placed in residential care. Similarly, the average cumulative time spent in
out-of-home care was 447 (std. 503) days, an average that increases for youth placed in family
Individual Factors Beta S.E. Wald Exp(b)
Age at initial placement .111 .012 93.59 1.118***
Male (female ref) .468 .038 153.1 1.597***
Behavioral problems 1.670 .044 1909.2 5.313***
Police .009 .057 .045 1.009
Request for youth criminal justice prior to
placement .756 .023 173.2 2.130***
Number of intestigations prior to placement .012 .020 .255 1.012
Socioeconomic disadvantage -.142 .169 49.98 .868***
Cox and Snell (R2) .208
* p < .05 ** p < .01 *** p < .001
Table 2: Logistic model predicting generalized propensity to placement in residential care
(N=15,518)
© Esposito, Chabot, Dalaye, and Trocmé
88-100
96 Copyright © 2015 International Journal of Child and Adolescent Resilience
A total of 1 placement change
Total
8,792
Event
3,720
No Move
5,072
%Event
42.3%
A total of 2 placement changes
Total
7,304
Event
2,232
No Move
5,072
%Event
30.5%
3 or more placement changes
Total
9,566
Event
4,494
No Move
5,072
%Event
46.9%
A total of 1 move A total of 2 moves 3 or more moves
Individual Factors Exp(B) (95% CI) Wald Exp(B) (95% CI) Wald Exp(B) (95% CI) Wald
Age at initial placements:
14-17 (10-13 ref) 1.085 (.961, 1.123) 1.74 1,307** (1.107, 1.544) 9.95 1.702*** (1,405, 2.062) 29.5
Gender:
Male (female ref) 1.030 (.935, 1.134) .359 .929 (.817, 1,057) 1.24 1.052 (.817, 1,057) .480
Reason for initial placement:
Psychological & emotional
abuse
Physical, material & health
neglect
Parent high risk lifestyle
School truancy & neglect
Risk of or sexual abuse
Behavioural problems
Risk of or physical abuse (ref)
1.052
.626*
.789*
1.230
.961
1.176*
(.807, 1,372)
(.431, .909)
(.660, .945)
(.945, 1.601)
(.740, 1.248)
(1.010, 1.370)
.140
6.04
6.66
2.37
.090
4.33
1.063
.233***
.696**
1.139
.775
1.308*
(.740, 1.527)
(.119, .455)
(.540, .896)
(.794, 1.633)
(.534, 1.124)
(1.057, 1.619)
.110
18.2
7.88
.498
1.80
6.07
.703
.291***
.505***
.821
.734
1.532***
(.448, 1.103)
(.153, .552)
(.376, .679)
(.535, 1.260)
(.479, 1.125)
(1.203, 1.951)
2.35
14.2
20.6
.814
2.01
11.9
Source of referral at initial
placement:
CLSC
Youth protection agency
Police
Other professional institutions
School
Hospital staff
Unidentified
Family (ref)
.889
1.181
1.064
.933
1.050
.972
.702
(.749, 1.055)
(.985, 1.415)
(.940, 1.204)
(.743, 1.172)
(.908, 1.215)
(.784, 1.206)
(.478, 1.031)
1.80
3.24
.975
.356
.433
.066
3.25
.878
1.102
1.104
.685*
.908
.797
1.122
(.698, 1.104)
(.867, 1.401)
(.936, 1.301)
(.493, .951)
(.741, 1.112)
(.591, 1.074)
(.728, 1.729)
1.24
.628
1.38
5.11
.875
2.22
.272
.776
1.133
1.308**
.807
.863
.507***
.923
(.599, 1.004)
(.866, 1.481)
(1.097, 1.560)
(.571, 1.142)
(.689, 1.081)
(.353, .729)
(.559, 1.523)
3.72
.830
8.91
1.46
1.64
13.4
.099
Time in out-of-home care 1.003*** (1.002, 1.003) 694 1.005*** (1.004, 1.005) 1216 1.007*** (1.006, 1.007) 2306
Initial placement in residential
care (family foster care ref) 1.154** (1.053, 1.265) 9.46 1.723*** (1.522, 1.951) 73.6 1.872*** (1.632, 2.148) 79.9
Request for youth criminal
justice services 1.378*** (1.225, 1.551) 28.3 1.490*** (1.277, 1.738) 25.6 1.457*** (1.233, 1.722) 19.5
Number of investigations 1.157*** (1.096, 1.222) 27.5 1.293*** (1.207, 1.384) 54.2 1.341*** (1.245, 1.444) 59.7
Socioeconomic disadvantages .990 (.943, 1.040) .148 1.017 (.952, 1.086) .256 .963 (.897, 1.035) 1.05
Cox and Snell (R2) .126 .280 .549
* p < .05 ** p < .01 *** p < .001
Table 3: Multivariate logistic regression with inverse propensity weights predicting placement for
youth aged 10 to 17 years.
97
Volume 1, Number 1, 2013, pp. Volume 3, Number 1, 2015, pp.
foster care versus those placed in residential care. The composite index of socioeconomic
disadvantages for youth placed in out-of-home care is .22 (std. 93), an estimate that increases
in socioeconomic disadvantages for youth placed in family foster care compared to those
placed in residential care.
Given the noted differences in youth placed in residential care compared to those
placed in family foster care, a propensity model was estimated using covariates, which are
conceptually and methodologically most related to differences for youth placed in residential
care versus family foster care (See Table 2). Table 2 presents the results of the multivariate
logistic regression model used to predict the propensity weights to residential care at initial
placement. Findings from the multivariate logistic model reflect those reported in part
by Farmer and associates (2008) and Park & Ryan (2009), showing significant positive
differences between increased age at initial placement, male gender, severe behavioral
problems, and youth crime. The multivariate propensity model produced a Cox and Snell
R2 of .208, indicating that close to 21% of the variance in placement in residential care is
explained by the combination of age, male gender, behavioral problems and youth crime. In
Quebec, youth manifesting severe behavioral problems are 431% more likely to be admitted
to residential care and 113% more likely if they committed a crime prior to initial placement.
Propensity findings also revealed that socioeconomic disadvantages statistically decreased
youth chances to placement in residential care. In other words, the more socioeconomically
disadvantaged the youth, the more likely they are to be placed in family foster care. This is
primarily because the increased risk of initial placement in foster family care is explained by
parent and family difficulties for which socioeconomic vulnerabilities plays a key role. These
estimates were saved for each youth and propensity weights were then computed.
Using propensity scoring as model weights, three multivariate models were constructed
to examine to the influence of covariates on placement changes. The results of the three
multivariate logistic models are reported in Table 3. An adjusted odds ratio of more than
1 indicates increased chances of changing placements a total of one time, two times, or
at least three times from initial placement. Accounting for differences in youth placed in
family foster and residential care, the increased likelihood of changing placements was
statistically explained by a combination of: older age; behavioral problems; police reporting;
longer time in out-of-home care; increased number of investigations; youth criminality;
and, residential care. While 14 to 17 year olds were statistically more at risk of experiencing
2 or more placement chances, age was not a significant predictor of changing placements
once. Similarly, police reporting predicted 3 or more placement changes, but not less. All
other mentioned covariates predict placement changes for older youth, from the first to last
placement change. Aside from a request for youth criminal justice services whose magnitude
of influence decreases with each placement change, all other mentioned covariates increase in
magnitude of influence as the number of placement changes increase. Also, factors associated
to neglect; specifically parents’ high risk lifestyle and physical, material, and health neglect
decreases the changes of experiencing multiple placement changes. Lastly, in predicting
© Esposito, Chabot, Dalaye, and Trocmé
88-100
98 Copyright © 2015 International Journal of Child and Adolescent Resilience
multiple placement changes, the final multivariate model produced a Cox and Snell R2 of
.549 indicating that close to 55% of the variance in multiple placement changes is explained
by the model; a R2 increase from .126 for the multivariate model predicting a total of one
placement change to .549 for the model predicting 3 or more placement changes.
Discussion
Using propensity weights to control for pre-placement differences between youth
placed in residential care and youth placed in foster care, this study found that residential
care is a significantly less stable placement setting than family foster care; youth in residential
care are close 87% more likely to experience three or more placement changes compared
to their counterparts in family foster care. While previous studies have found a similar
association between residential care and placement instability, youth placed in residential
were considered to be inherently at greater risk of placement breakdown because of their
behavioural profile. Propensity score matching provides a level of statistical control similar
to random assignment, such that the greater likelihood of placement disruption in this study
can be attributed to the placement rather than to pre-placement differences.
Also, consistent with factors reported in Esposito et al (2014), and to a lesser degree
Farmer et al. (2008), Park & Ryan (2009), Connell et al. (2006), and Ossterman et al.
(2007), and contrary to those reported by James et al. (2004), this study finds that older
youth admitted to out-of-home care with behavioral problems, youth criminality, increased
number of investigations, and placed in residential care are the most likely to experience
multiple placement changes. Contrary to Esposito et al. (2014), controlling for differences
in the profile of youth placed in residential care versus family foster care, socioeconomic
disadvantages did not significantly explain placement changes. In line with James et al.
(2006), this study also found that the longer youth are placed the more chances they have
to experience a change of placement. Together in a cumulative fashion, youth aged 14 to
17 years old (odds, 1.702), admitted to residential care (odds, 1.872), placed because of
severe behavioral problems (odds, 1.532), reported by the police (odds, 1.308), and with
a confirmed act of delinquency (odds, 1.457), are the most likely to experience placement
disruption than any other youth in out-of-home care.
Practice implication
Residential care is the default out-of-home placement option for older youth placed
in child protection care in Quebec; close to 80% of older youth who enter out-of-home care
in Quebec are placed in residential care or end up being moving to residential care from
family foster care. While for some youth residential care may in fact be the best placement
option, findings from this study show that youth experience far less instability when placed
in family foster care. The analytic methods used in this study show that the greater stability
of family foster care could not simply be attributed to differences in the youth placed in foster
99
Volume 1, Number 1, 2013, pp. Volume 3, Number 1, 2015, pp.
care compared to residential care. In fact, many of the youth placed in residential care have
similar pre-placement profiles to youth entering family foster care. In light of these results,
far more efforts should be made to develop family foster care alternatives for older youth
being placed in care in Quebec.
Limitations
While the methodology is unique in allowing for a quasi-randomization treatment of
residential and family foster care, it is not without limitations. One such limitation is that
the clinical-administrative data underestimates the prevalence of placement changes as
youth informal placement with kin are not captured here. Second, the study did not adjust
for correlations that may exist because of siblings, nor is it able to unduplicated cases across
child protection jurisdictions. A longitudinal replication balancing the data based on the
propensity scores in order to match youth placed in residential care to family foster care –
reducing the clinical population studied by an estimated 60% - may help confirm the strength
and directions of estimates reported in this analysis. Also following the work of Ryan and
Park (2008), further analysis will examine whether placement settings itself influences later
youth crime using a propensity matched clinical population of youth admitted to family
foster and residential care in Quebec.
References
Barth, R. P., (2002). Institutions vs. foster homes: The empirical base for a century of action. Chapel Hill, NC: UNC,
School of Social Work, Jordan Institute for Families.
Connell, C. M., Vanderploeg, J. J., Flaspohler, P., Katz, K. H., Saunders, L., & Tebes, J. K. (2006). Changes in
Placement among Children in Foster Care: A Longitudinal Study of Child and Case Influences. The Social
service review., 80(3), 398.
DeGue, S., & Spatz Widom, C. (2009). Does out-of-home placement mediate the relationship between child
maltreatment and adult criminality? Child Maltreatment, 14(4), 344-355.
De Swart, J. (2011). De professionele jeugdzorgwerker [The professional youth care worker]. Enschede: Saxion
(Dissertation).
De Swart, J. J. W., Van den Broek, H., Stams, G. J. J. M., Asscher, J. J., Van der Laan, P. H., Holsbrink-Engels, G. A.,
et al. (2012). The effectiveness of institutional youth care over the past three decades: A meta-analysis.
Children and Youth Services Review, 34, 1818–1824.
Esposito, T., Trocmé, N., Chabot, M., Shlonsky, A., Collin-Vézina, D., & Sinha, V. (2014). The stability of child
protection placements in Québec, Canada. Children and Youth Services Review 42, 10-19.
Farmer, E. M. Z., Mustillo, S., Burns, B., & Holden, E. (2008). Use and Predictors of Out-of-Home Placements
Within Systems of Care. Journal of Emotional and Behavioral Disorders, 16(1), 5-14.
Guo, S. & Fraser, M. (2014). Propensity score analysis: Statistical methods and applications. Sage: Thousand Oaks,
CA.
Harder, A. T. (2011). The downside up? A study of factors associated with a successful course of treatment for
adolescents in secure residential care. Groningen: Rijksuniversiteit Groningen (Dissertation).
James, S., Leslie, L., Hurlburt, M., Slymen, D., Landsverk, J., Davis, I., . . . Jinjin, Z. (2006). Children in Out-of-
Home Care: Entry Into Intensive or Restrictive Mental Health and Residential Care Placements. Journal of
Emotional and Behavioral Disorders, 14(4), 196-208.
Knotter, M. H.,Wissink, I. B.,Moonen, X. M. H., Stams, G. J. J. M., & Jansen, G. J. (2013). Staff’s attitudes and
© Esposito, Chabot, Dalaye, and Trocmé
88-100
100 Copyright © 2015 International Journal of Child and Adolescent Resilience
reactions towards aggressive behaviour of clients with intellectual disabilities: A multi-level study.
Research in Developmental Disabilities, 34, 1397–1407.
Leathers, S. J. (2006). Placement disruption and negative placement outcomes among adolescents in long-
term foster care: The role of behavior problems. Child Abuse & Neglect, 30(3), 307-324. DOI: http://dx.doi.
org/10.1016/j.chiabu.2005.09.003
Loseke, D. R., Gelles, R. J. et Cavanaugh, M. M. (2005). Current controversies on family violence. Thousand Oaks:
Sage.
O’Brien, R.M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality and Quantity,
41(5), 673-690.
Oosterman, M., Schuengel, C., Slot, N. W., Bullens, R. A.R., & Doreleijers, T. A. H. (2007). Disruptions in foster
care: A review and meta-analysis. Children and Youth Services Review, 29, 53–76.
Park, J. M., & Ryan, J. P. (2009). Placement and Permanency Outcomes for Children in Out-of-Home Care by
Prior Inpatient Mental Health Treatment. Research on Social Work Practice, 19(1), 42-51.
Ryan, J.P., Marshall, J., Herz, D., & Hernandez, P. (2008). Juvenile delinquency in child welfare: Investigating
group home effects. Children and Youth Services Review, 30, 1088-1099.
Souverein, F.A. Van der Helm, G.H.P & Stams, G.J.J.M. (2013). ‘Nothing works’ in secure residential youth care?
Children and Youth Services Review, 35, 1941-1945.
Thoemmes, F. (2012). Propensity score matching in SPSS. Department of Human Development, Cornell
University, Ithaca. Retrieved December 11, 2015: http://arxiv.org/ftp/arxiv/papers/1201/1201.6385.pdf
