Children and Youth Services Review 34 (2012) 2208–2219

Contents lists available at SciVerse ScienceDirect

Children and Youth Services Review

journal homepage: www.elsevier.com/locate/childyouth

Attachment theory and change processes in foster care

David J. Tucker a,⁎, Michael J. MacKenzie b
a
School of Social Work, University of Michigan, United States
b
School of Social Work, Columbia University, United States

a r t i c l e i n f o a b s t r a c t

Article history: Despite wide acceptance in the multifaceted field of child care policy and practice, attachment theory has
Received 4 June 2012 found limited use in examining empirically the circumstances and conditions of special populations of
Received in revised form 24 July 2012 children. This inquiry addresses this limitation by elaborating attachment theory as a foundation for
Accepted 30 July 2012
contemporary foster care practice and policy. We focus on how caregiving contexts and the nature of their
Available online 9 August 2012
change selects certain characteristics and behaviors as relevant in explaining a child's risk of placement
Keywords:
change in, or exit from, foster care. We use data on a population of 3448 foster children over a 21-year period
Attachment theory to test arguments that children's strategies for dealing with change can be both resistant and adaptive, and
Behavioral inertia that self-perpetuating patterns of attachment can contribute to increasing rates of change in children's lives.
Foster care Results strongly support attachment theory as a transactional theory of change. Placement change not only
Placement change influences the hazard of exit in the manner predicted but also engenders a “liability of change,” with early
Internal working model change influencing the likelihood of future change independent of contextual and child characteristics. From
the perspective of this inquiry, future research that omits information on the history and timing of significant
changes in children's lives will be limited in its capacity to explain their current circumstances.
© 2012 Elsevier Ltd. All rights reserved.

1. Introduction hypotheses designed to test AT's efficacy as an interpretive lens for

foster care dynamics; and then explore in detail the results and their
1.1. Purpose implications.

The relevance of insights from John Bowlby's attachment theory (AT)
to child care policy and practice are so widely accepted that the need for
confirmation through empirical testing and analysis is only cursorily
acknowledged. Significantly, such acceptance has helped promote two
complimentary limitations (Rutter & O'Connor, 1999). One is a tendency
to apply the principles of AT across differing populations of children
without qualification. The complement is its limited use in examining
the circumstances and conditions of special populations of children by
elaborating relevant aspects of the theory in terms specific to such
populations.
This inquiry seeks to address these limitations by exploring AT as a
foundation for contemporary foster care practice and policy. We focus
mainly on how caregiving contexts and the nature of their change
selects certain characteristics and behaviors as relevant in explaining
a child's risk of placement change in, or exit from, foster care. Within
this framework, we seek to determine, empirically not cursorily,
whether AT can deepen understanding of the transactional nature of
contemporary foster care dynamics. To this end, we review recent
work in the research tradition of AT and in foster care; posit several
⁎ Corresponding author.
E-mail address: celdjt@umich.edu (D.J. Tucker).
1.2. Background
In foster care practice, Bowlby's insights undergird a strong
professional consensus regarding the importance of matching foster
home strengths and capabilities with child need and pursuing quick-
as-viable family reunification (Pecora, Whittaker, Maluccio, Barth, &
DePanfilis, 2009; Steinhauer, 1991). Both practices are regarded as
important in minimizing potentially negative effects of relationship
disruption on children (Rutter & O'Connor, 1999). Failing reunifica-
tion, the consensus calls for creating conditions that AT predicts are
crucial to a child's healthy psycho-social development, stable and
secure social bonds with significant adult caregivers (Bowlby, 1988;
D'Andrade, 2005; Steinhauer, 1991).
In practice, foster care dynamics seldom yield conditions empirically
consistent with the implications of AT principles. Indeed, evidence of
constant-to-increasing duration times to foster care exit or family
reunification indicate that the potentially negative effects for children of
relationship disruption are not sufficiently minimized (Child Welfare
Information Gateway, 2010; Children's Bureau, 2010). Findings also
point to extended duration in care without movement to permanency,
and of comparatively frequent placement changes. Both support an
expectation of proportionately higher numbers of children with behavior
problems (Children's Bureau, 2010).

0190-7409/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.childyouth.2012.07.020
 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
Motivated by these stubborn realities, researchers are working to
identify relevant causal factors and conditions; recent studies show
that such factors as age and nature of placement setting can predict
both duration to exit and risk of placement change (Akin, 2011;
Connell et al., 2006; van Santen, 2010). Researchers thus suggest that,
if the effects of such factors can be mitigated, greater stability and
more continuity in child-caregiver relationships will result, thereby
producing more positive outcomes for children (Connell et al., 2006;
Pecora et al., 2009; Rutter & O'Connor, 1999).
The research reported in this paper builds on and contributes to
this literature. However, our emphasis is not solely on examining
relationships between rates of placement change or exit and child
characteristics. We are also concerned with the fact that relationships
between child characteristics and rates of change and exit affect
variation in the prevalence of child characteristics themselves. That is,
while current research treats child-specific variables like age as
influencing rates of change and exit, distributions of such character-
istics in observed populations also reflect effects of earlier changes in
rates of change or exit (Swan & Sylvester, 2006).
Thus, it is vital to understand how child characteristics influence risk
of placement change or exit as they interact with the caregiving
contexts in which foster children are embedded. Accordingly, this study
focuses less on how particular characteristics affect a child's propensity
to exit care or sustain placement change. Rather it generates and tests
hypotheses about dynamic transactional processes underlying how
caregiving contexts select certain characteristics and behaviors (includ-
ing the process of change itself) as important in explaining a child's risk
of change or exit.
1.3. Views of the child in attachment theory
As a “general theory” or “research tradition” (Laudan, 1977), AT has
spawned a range of specific and sometimes conflicting ideas about how
to answer questions and solve problems in the intersecting domains of
child development and behavior.
One set of conflicting assumptions—those regarding children and
change—are particularly important to this investigation. Some represent
children as malleable, able to adapt to the requirements of change, given
a supportive contemporaneous environment (Lewis, 2001). Others
understand them as inflexible, based on arguments that attachment
patterns tend to be self-perpetuating and change resistant regardless of
context (Thompson & Raikes, 2003). The first view implies the viability
of studying children's behavior using cross-sectional designs and static
forms of analysis. The second emphasizes children's development
histories and affirms longitudinal designs and dynamic forms of
analysis. Following others (Bowlby, 1969; MacKenzie & McDonough,
2009; Sroufe, Egeland, & Kreutzer, 1990), we construe these views as
complimentary and adopt Bowlby's internal working model concept
as an organizing assumption (Bowlby, 1969, 1973, 1980). Thus, we take
into account both development history and current circumstance.
An internal working model—a relational network of internalized
knowledge constructed on the basis of attachment experiences—
encapsulates a child's expectations of external reality and its role in
that reality. The child uses its internal working model to anticipate
and evaluate probable outcomes of purposive relationships, establish-
ing the security needed to form relationships. In time, cognitive
repertoires associated with internal working models are routinized
(Bowlby, 1973; Crick & Dodge, 1994), and the child requires less
conscious deliberation to invoke the working model. Thus the child
manages relationships more stably and efficiently.
From the internal working model perspective, children approach
new situations not as blank slates but with “…certain preconceptions,
behavioral biases, and interpretive tendencies” (Sroufe, Carlson, Levy, &
Egeland, 1999, p. 5) and tend to interpret current environmental cues in a
manner consistent with past interpretations (Dodge, Pettit, Mcclaskey, &
Brown, 1986). They also elicit reactions from the environment consistent
2209
with maintaining, rather than changing, existing behavioral reper-
toires (Bowlby, 1973; Bretherton & Waters, 1985; Crick & Dodge,
1994; Sroufe et al., 1999). This implies differential survival rates for
children's relationships with caregivers. On one hand, failure to
adapt by either or both parties could result in termination of a
relationship (Fichman & Levinthal, 1991). On the other, unilateral or
mutual adjustments could increase chances of relationship success.
This leads to the expectation that patterns of behavior evinced by
children in dealing with contexts of relationship change will reflect both
resistant and adaptive tendencies and that, in fact, self-perpetuating
patterns of attachment can actually contribute to increased rates of
relationship change in children's lives.
In this paper, we use the working model approach to children and
change to derive seven hypotheses particularized to children in
foster care: the first three addresses how placement change affects
risk of exit from foster care; the remaining four consider the effects
of change itself on subsequent rates of placement change, indepen-
dent of the effects of contextual and child characteristics. While the
working model idea is owed to Bowlby, we expand its scope by
incorporating aspects of representational theory, theory of mind,
transactional theory, and organizational ecology. The result is a broader
conception of children's working models which, combined with our
analysis, affirms the importance of conditioning interpretations of AT to
the conditions and circumstances of special populations of children.
2. Hypotheses
2.1. Placement change and risk of exit
Recent research on internal working models identified and described
their internal components, or schemas (Bretherton & Munholland, 1999;
Bretherton & Waters, 1985; Main, Kaplan, & Cassidy, 1985). These
sequentially ordered structures contain information about how one
interacts with specific others, under certain conditions and in specific
situations, to achieve particular ends (Schank & Abelson, 1977). Schemas
define a child's knowledge of relationships and the actions and outcomes
associated with the effectuation of that knowledge. While the idea of
working models as organized bundles of schemas implies differences
among children's working models, current literature identifies a limited
number of major forms of attachment organization (Crittenden, 1999;
Main et al., 1985) as the range of possible caregiver responses to
attention-seeking behavior also is limited (Main et al., 1985). However,
the idea of schema-based working models does give a more precise basis
for thinking about how change will affect a child's relationship with its
environment. Specifically, if stable, reproducible schemas are integral to
a child's attachment behavioral system and guide behavior, and if change
calls into question the efficacy of these schemas, change is likely to
produce anxiety, defensive exclusion, and defensive misattribution that
interfere with the child's ability to contend constructively with its
immediate environment (Bretherton & Munholland, 1999). Within the
foster care system, these behaviors are likely to prove disruptive,
lowering a child's chances of adapting well to a caring context and
increasing the chances of being labeled problematic. This in turn likely
will lower the possibility, or risk, of exiting the foster care system. Thus:
Hypothesis 1. Placement change in foster care decreases a child's risk
of exiting foster care, independent of the effects of other character-
istics and conditions.
Hypothesis 1 implies the effects of placement change are uniform
across children. We think this is unlikely in one important sense.
Specifically, we expect child's age at time of placement change will
affect subsequent risk of exit. How that effect will manifest is subject
to competing arguments.
The first argument is consistent with Bowlby's claims (Bowlby,
1969, 1973, 1988) that a positive relationship exists between increasing
age and increasingly inflexible patterns of behavior, and with Sroufe's
2210 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
(1997) assertion that pathology is “… a pattern of adaptation reflecting
the totality of the developmental context to that point.” Applied to foster
care, this argument implies that as children age into foster care, patterns
of behavior are likely to become less flexible, resulting in more consistent,
more apparent patterns of dysfunctional attachment or detachment
behaviors. In turn, such behaviors are likely to lead to assessments that
they remain in care. Overall, therefore, this argument suggests that the
disruptive effects of change will increase with age. Thus:
Hypothesis 2a. The older the child at the time of a placement change,
the lower the subsequent risk of exit.
Crittenden's (1999, 2000) dynamic-maturational approach to attach-
ment suggests an alternative argument. Specifically, it cautions against
overemphasizing security and underemphasizing danger as children's
basis for working models. Danger has been a constant feature of human
environments, and it is reasonable to presume that, as a matter of
survival, children construct self-protective strategies to deal with it.
These strategies are designed to influence caregiver behavior, rendering
it more predictable to the child in self-protective ways. For younger
children, these strategies are simple and direct. However, with
maturation, children are increasingly able to consider and modify
their own mental processes, resulting in more sophisticated and
complex strategies characterized, among other things, by a “freeing
of mental activity from behavioral enactment” (Crittenden, 1999, p.
169). For example, “… a school age child may mislead others with
regard to his intentions by hiding evidence of his mental function-
ing,” or an adolescent may hide her feelings with one person but act
on them with another (Crittenden, 1999).
Assuming that foster children generally consider themselves more
vulnerable when living among strangers ( Lyon, 1987; Mrazek & Mrazek,
1987; Steele, 1986; Weinstein, 1960), children aging into foster care will
represent themselves behaviorally as adapted to that environment in the
interest of increasing their chances of exiting foster care and returning
home, or living independently. When presented with a change, they
could show themselves as coping well, thus increasing the prospects of
exiting. Empirically this manifests in a direct relationship between age
and risk of exit, suggesting:
Hypothesis 2b. The older the child at the time of a placement change,
the higher the subsequent risk of exit.
For children persisting in foster care, effects of placement change are
not likely to disappear immediately. Changes prompt circumstances and
conditions that parallel repeated attempts to initiate establishment of
new relationships. Both theory and research (Jovanovic, 1979; Land,
1984; March & March, 1977; Tucker, Hurl, & Ford, 1994; Vuchinich,
Teachman, & Crosby, 1991) hold that relationships have the highest risk
of failure during initial formation, with risk of termination declining
monotonically. That is, the longer a relationship lasts, the less likely it is to
terminate. Given a degree of environmental predictability after change, a
person can rebuild and reorganize internal structures and processes,
stabilizing capacity to develop and maintain external relations (Bowlby,
1988).
For foster care, this implies that the longer a child's relationship
with a foster family, the child is more likely to fit with that context
and less likely to be labeled problematic. Hence, we expect stability in
placement after change to be associated with increased risk of exit.
Thus:
Hypothesis 3. Disruptive effects of placement change attenuate with
length of time since last change, resulting in a higher propensity to
exit the system.
2.2. Behavioral inertia and placement change
According to the conventional conception of working models in
AT, how children deal with changes in social relationships in current
care-giving environments depends, in part, on how secure similar
relationships were in earlier environments. Secure attachments' positive
effects on confidence and self-image promote capacity for exploratory,
autonomous behavior vis-à-vis current environments (Verschueren,
Marcoen, & Schoefs, 1996). Children with secure earlier attachments
thus have potential for greater sensitivity, or openness, to the specifics of
any new context. This implies that these children have a higher
probability of experiencing change and more opportunities to learn
about its meaning and management. Insecure attachments contribute
to lower levels of confidence and a poorer self-image and thus, to less
openness to the specifics of a given context and more timid approach to
engaging it. Accordingly, children with insecure attachments who
experience change in a care-giving environment will experience it
mainly as a negative external imposition, something to be resisted for
the costs it exacts, not embraced for the benefits it bestows. It is
reasonable that children with secure past attachments are more likely
to adapt to the requirements of a new situation than those with a
history of insecure attachments. Hence they are likely to have lower
rates of change.
The capacity to adapt to new environments is not without limit.
Indeed, we argue that children tend to evolve relatively stable
strategies of environmental engagement and are reluctant to change,
even when confronted with contexts for which these strategies may
not be highly adaptive. Clearly, a child's behavioral repertoire cannot
be so flexible that it anticipates the requirements of adapting to the
full range of environmental contingencies to which it may be
exposed. By the same token, a one-size-that-fits-all strategy that
depends on environmental constancy is not likely to function either.
It is more plausible that optimal environmental engagement re-
quires a balance between acknowledging the environment's com-
plexity and treating it as an unchanging constant (Bowlby, 1969,
1973). This strategy enhances efficiency and complexity. Based on
past learning and experience, a child's behavior patterns become
organized and consistent, strengthening its security and competence
in uncertain environments. However, this strategy also promotes
behavioral inertia. Because routinized behaviors relieve a child of the
requirement to deal with each social interaction by conscious learning
and decision-making, routines are efficient and valuable. Hence, a child
may meet with resistance circumstances that need or demand behavioral
change (Crick & Dodge, 1994, pp. 80–81).
Assuming a propensity to resist change requires attention to the
timing of changes. Prospects for adapting to change improve if a child's
response matches the temporal pattern of change in its environment.
However, our argument for behavioral inertia implies that this is
unlikely, suggesting that behavioral change tends to be slow relative to
environmental change. Behavioral inertia is indicated when a child's
adjustment lags substantially behind the rate of environmental change.
Because of their greater environmental sensitivity and superior behav-
ioral repertoires, we expect children with more secure attachment
histories to demonstrate lower levels of behavioral inertia vis-à-vis
environmental change—adjusting more quickly, thus successfully to
caregiving change—than children with insecure attachment histories.
Foster care is often characterized by instability in care-giving
environments brought about by frequent changes in children's place-
ments. Considering behavioral inertia, few things seem as adverse to a
child's wellbeing as the child committing to change only to find, upon
changing, that the context requires different behavior! How is a child
likely to react? How will he or she organize, behaviorally, to deal with
subsequent care-giving environments?
We think the answer lies in recognizing that, when it comes to foster
care, the propensity toward behavioral inertia is different, in that there
are twin implications for change. As described above, behavioral inertia
occurs in a conventional sense when a child maintains a particular
pattern of behavior and resists change. But, there is another form of
inertia, momentum (Amburgey, Kelly, & Barnett, 1993), which occurs
when the processes invoked for dealing with change are themselves
 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
routinized. Applied in the context of this research, the momentum view
of inertia allows that the way a child dealt with a particular kind of
change in the past may be the way he or she manages it in the future.
Thus, the child uses change to deal with problems in the environment,
regardless of whether it solves those problems (Levitt & March, 1988).
Concomitantly, the child's own behavior becomes a factor in explaining
change (Sameroff & MacKenzie, 2003).
Building on our inertia-momentum argument, we turn to Bowlby's
(1973) pathways model of personality development. Represented
metaphorically as a train on a main line switching to branch lines, it
elaborates how a child's prior experience with change may produce
maladaptation in his or her current environment (Bowlby, 1973, p. 365,
370). Specifically, maladaptive behavior emerges as successive “junction
switches” that take a child farther from pathways leading to competent
social functioning. Sroufe (1997) affirmed this argument, observing that
maladaptive behavior emerges from a succession of deviations from
normative patterns. Each succeeding change lowers the child's proba-
bility of adapting positively to subsequent contexts (see also Crick &
Dodge, 1994; Moffitt, 1993; Newton, Litrownik, & Landsverk, 2000).
In foster care, we expect this pattern to present in a child's
increasingly detached behavior, treating caregivers as members of a
singular, non-responsive category the child regards as not interested
in, or committed to, social interaction that will advance positive
outcomes for him or her (Howes, 1999). Caregivers may experience
reciprocal effects, increasingly regarding children with frequent
placement changes as inherently unmanageable and unrewarding
because of their behavior and (perceived) limited prospects for
regaining positive, adaptive developmental trajectories. With this
mutual skepticism about the benefits of investing time, effort, and
energy in relationship, child and caregiver commitments to relationship
become increasingly superficial; invoking placement change to manage
problems of social interaction becomes easier and may be routinized in
the child's behavioral repertoire. In this way, once a child experiences
placement change, momentum for additional changes may develop
since change is part of what the child expects, knows how to do, and
may even work to invoke (Newton et al., 2000; Palmer, 1996). Thus:
Hypothesis 4. The probability of a placement change increases with
the number of previous placement changes.
2.3. The dynamics of placement change
Hypothesis 4 implies something like a liability of change: Change
engenders change and thus a propensity not to persist in a given
context.
Yet, evidence indicates that children can move through periods of
anxiety, change, and maladaptive behavior to achieve periods of
relative stability (Horwitz, Balestracci, & Simms, 2001; Howes &
Segal, 1993; Lewis, Feiring, & Rosenthal, 2000; Weinfield, Sroufe, &
Egeland, 2000). Can this literature be reconciled with Hypothesis 4?
We think the answer may lie in the dynamics of change itself. More
specifically: the chance of a given change recurring depends on how
recently that type of change occurred. Foster care contexts provide
different treatment over time than care received in an earlier context.
Thus a child's working model of attachment relationships is delayed but
nonetheless gradually updated and reorganized. This reorganizing
results in the foregrounding and perhaps generalization of more
recently developed and used schema. Assuming all other things as
equal, and accepting that children are more likely to invoke recently
used simulations of reality afforded by newer schemas than older and
less frequently used ones, we argue that the longer the time since the
last invocation of schemas promulgating change as an adaptive
strategy, the less likely they are to be accessed and used in the
present. In foster care, the child's caregivers are likely to develop
routines that acknowledge and deal with the child's presence, both
shaping and conforming to the child's expectations of a working
2211
relationship, and conveying a view of the child as remaining in the
family (Tucker et al., 1994). Thus:
Hypothesis 5. The probability of placement change decreases with
the time since the last placement change.
Along with the dynamics underpinning Hypothesis 5, a child's age
also is likely to affect placement change. We expect that older children
will be more likely than younger children to experience placement
change. As children age they are treated differently by their environ-
ments: in many ways, these environments become more complex and
demanding. Since environmental change requires children to change
and adapt (Marvin & Britner, 1999), the nature and complexity of their
internal working models must change if they are to continue as
appropriate guides for the assessments of, and expectations in, given
social interactions. For these change to occur, a child must have available
sufficient attentional resources to acknowledge, interpret, and encode
relevant socio-behavioral information from its environment (Crick &
Dodge, 1994; Dodge et al., 1986; Macrae & Bodenhausen, 2000). In turn,
AT implies that, ceteris paribus, a child's attentional resources at any
given time are a function of current and past felt security.
By definition, foster care entails disruptions in family-based
caregiving arrangements and placement in unfamiliar environ-
ments. Moreover, many children have been exposed to problematic,
perhaps frightening, conditions in family caregiving contexts (Dozier,
Albus, Fisher, & Sepulveda, 2002). This can have damaging consequences,
including deficits in children's abilities to conceptualize, understand, and
make inferences about the thoughts, beliefs, and emotions of others
(Cicchetti, Rogosch, Maughan, Toth, & Bruce, 2003; Pears & Fisher, 2005).
Hence many children may enter foster care with limited attentional
resources, primed to develop, if not already evidencing, maladaptive
behaviors. Accordingly they may be limited in their capacity to develop
new and different relationships in current caregiving environments
(Sroufe & Fleeson, 1986) and may have problems negotiating subsequent
developmental issues and changes (Crittenden, 1999; Sroufe, 1997). As
such, comparatively older foster children may emerge as problems for
caregivers (James, 2004; Rhodes, Orme, & Buehler, 2001), which in turn
negatively influences foster families' ability or willingness to care for
them. Thus:
Hypothesis 6. The probability of a placement change increases with
the age of the child.
Finally, we think age also is consequential vis-à-vis the timing of
placement change. Bowlby's pathways model (1973), refined by Sroufe
(1997), conceptualizes children's problems not as specific conditions
endogenous to individuals but as maladaptations resulting from de-
viations over time in successive adaptations to the environment. As such,
it anticipates earlier failures to adapt as predictive of later failures
(Sroufe, 1997, p. 253). Considered in the context of our preceding
arguments, this proclivity implies that placement change probability
depends on change frequency, and how recently it occurred, and on its
timing in a child's tenure in foster care. That is—accepting placement
change as certainly stressful for a child and as indicative of some level of
failure to adapt—the implication of the pathways model, ceteris paribus,
is that early change is likely to be followed by a propensity for further
changes. Thus:
Hypothesis 7. An early change in placement increases the probability
of subsequent placement changes.
3. Methods
To test the seven hypotheses discussed above, we analyzed foster
child exits and placement changes for 3448 children taken into care in
a major urban area in Ontario, Canada, between January 1970 and
June 1990. Data sources were archival, and included confidential files
on the children maintained by the local child welfare authority as well
2212 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
as various lists, documents, and reports from other sources—including
different levels and departments of government, local agencies, and other
special-purpose child welfare bodies. These data sources provided time-
varying information on both children and their environments. For each
child, this included crucial timing information, recorded as to day, month
and year, on their entry into foster care, all placement changes, and their
final exit from the foster care system. We used these data to construct
event histories for each child in the form of a sequence of time periods, or
“spells,” separating each event. Each child's first spell began when he or
she first came into care and ended with the first change event the child
experienced—whether a placement change, meaning movement into a
new foster home, or an exit, meaning a termination of the child welfare
authority's responsibility to care for the child in a licensed foster care
facility. If the child persisted in foster care after the first change event, the
second spell began the day after the first event and ended with the
second event, and so on. We divided spells persisting for more than one
year into annual increments ending with a dummy event recording
whether or not it involved a change in placement status. This method of
organizing data allowed variable values to change over time and ensured
that time-varying covariates were updated annually.
Fig. 1 describes foster home exit levels by six-month periods. It
shows an initial low rate of exit, increasing to a relatively stable level by
the second half of 1971, dropping sharply in the second half of 1984 and
first half of 1985. After this, exit levels compare to the pre-1984 period,
coinciding with labor unrest in the child welfare authority that entailed
a protracted strike. Thus, we include a period effect in our exit models to
control for low exit rates during this time.
Fig. 2 describes the number of foster home placement changes by
six-month periods between 1970 and 1990. Prior to 1985, the overall
pattern resembles the exit pattern—an increase in 1971 and relative
stability until the 1984 decrease that corresponds the protracted
strike. Thus, we include a period effect in change models to control for
fewer placement changes. From 1985 until 1990, however, placement
changes show a constant pattern of increase, a trend we explore in
the discussion section.
3.1. Independent variables
We constructed a duration “clock” variable, rescaled by dividing by
100 to avoid unwieldy numbers. It records elapsed time (in days) since
last placement change. It is 0 until a change, at which point it records
the elapsed time since the change. If another change occurs, the clock is
reset to 0 and again records the elapsed time. These values were used to
evaluate Hypotheses 1, 3, and 5. Hypotheses 1 and 3 predict that the
duration clocks will have significant positive coefficients. A placement
change resets the time clock to 0, resulting in a drop in the hazard value
20 40 60 80 100 120 140 160 180 200
Numbers of Exits
0
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990
Year
Fig. 1. Number of foster home exits by six-month intervals 1970–1990.
100 200 300 400 500 600 700 800
Numbers of Changes
0
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990
Year
Fig. 2. Number of placement changes by six-month intervals 1970–1990.
and thus an increased risk of exiting the foster care system, as anticipated
by Hypothesis 1. The passage of time results in an increasing hazard and
thus a higher risk of exiting the system, as anticipated by Hypothesis 3.
On the other hand, Hypothesis 3 predicts that the duration clock will
carry a significant negative coefficient; after a change, the hazard of
change increases initially but then declines over time.
Evaluating Hypotheses 1 and 3 requires us to distinguish between
the effects of changing states and the effects of being in a specific state
(Amburgey et al., 1993). For this reason, we used gender to capture
the state-specific effect of being either male or female (Markovits,
Benenson, & Dolenszky, 2001).
A dummy variable records whether or not a placement change
occurs, coded 0 until a change occurs and 1 thereafter, regardless of the
number of changes. Accordingly this variable captures the average
effect of change, net of other factors (Amburgey et al., 1993). To test
Hypotheses 2a, 2b and 7, we interacted this variable with the child's age.
The interaction term takes a value of 0 prior to the change and the
child's age after that.
Hypothesis 2a predict that the interaction term will have a significant
negative coefficient, implying a decreasing hazard of exit, whereas
Hypothesis 2b predicts a significant positive coefficient, implying an
increasing hazard of change. For Hypothesis 7, the interaction term
contributes nothing to the hazard of change prior to a placement change.
If a change occurs early in a child's foster care tenure, the child's age will
be lower, resulting in lower values for the hazard of change. By the same
token, if the first change occurs later in a child's tenure in foster care, age
is higher and the effects on the hazard of change greater. Conversely, for
Hypothesis 2a and 2b, if the first change occurs earlier in a child's tenure
when the age is lower, the effects on the hazard of exit will be greater. If
the first change occurs later when age is higher, the effects on the hazard
of exit will be less.
We used data on children's chronological ages, calibrated in days, to
test Hypothesis 6. Similar to the duration clock variable, we rescaled age
by dividing by 100.
Finally, to evaluate Hypothesis 4, we created a variable counting
the cumulative occurrences of placement changes. This variable has a
value of 0 until the first change, of 1 until the second change, 2 until
the third change, and so on.
3.2. Control variables
We incorporated a number of realistic but otherwise extraneous
control variables in our models to evaluate whether or not our
hypotheses are supported despite the presence of other plausible
sources of explanation. Two variables—local unemployment and local
labor force participation rates of women—refer to general economic
 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
conditions that may have implications for foster care dynamics. Women
historically have been the main source of childcare labor in families.
Accordingly, their increased employment outside the home without
simultaneous change in the distribution of responsibility in the home
may engender conditions that dampen the return of children to their
own homes. Women's increased out-of-home employment may also
create conditions supportive of increased movement of children from
one foster care placement to another. On the other hand, women's place
in the economy is more tenuous than men's. Hence, in times of increased
unemployment they are more likely to be fired or laid off, thereby
possibly making more labor available to care for children. The
possible consequences include increased foster care exit rates and
lower placement change rates.
We included two control variables measuring possible effects of
policy shifts aimed at instituting economic incentives and institutional
change in the child welfare authority. Each was treated as a historical
event, coded 1 when the policy is in effect and 0 otherwise. Each was
intended to have, and can be construed as having, effects on the exit and
change rates of children in care. Economic incentives refers to a period
beginning in 1973 when foster home rates were increased, and policies
aimed at establishing fostering as an economically rewarding career
option were introduced. Institutional change refers to a period beginning
in 1978 when formal legislative expectations concerning relationships
among foster homes, foster children, and the agencies were changed
such that the rights and needs of foster children achieved increased
priority against the collective rights and needs of foster homes. Further,
foster homes were constrained by these changes to behave less as
autonomous self-directing family units and more as partners, cooperat-
ing with child welfare staff in setting goals and implementing plans for
the children in care (Tucker & Hurl, 1992).
Our next two variables refer first to the period of labor unrest and
second to a method of foster care practice. Labor unrest was measured
as a period effect, coded 1 when in effect and 0 otherwise. The method
of practice, matching, emphasized the importance of workers making
every effort, when placing children in foster homes, to ensure a good fit
between a child's needs and foster home characteristics and capabili-
ties. We measured this indirectly as the ratio in any given year of the
numbers of children in care to the number of licensed foster homes.
There are conflicting arguments about the possible effects of this
ratio. On the one hand, it can be assumed that a high ratio, indicating
proportionately more children than foster homes, means fewer choices
for workers and thus the potential for poor matches, resulting in lower
propensity to exit and a higher propensity to change. Alternatively, it
can be assumed that a high ratio implies more control by foster homes
over what child they will accept into their care. Hence, they are more
invested in the decision, once it is made, resulting in a lower propensity
for children to change placements and a higher propensity for them to
exit.
Our final control variable is overall duration dependence, or the
possible effect of total time spent in care on a child's hazard of change
or exit at any given point. We include this variable because general
models are exponential models and therefore do not assume a
particular functional form for the underlying pattern of duration
dependence. We would argue that in the case of the exit from care or
changes in placement, this likely is not correct. The least problematic
children will likely exit earlier; over time proportionately more
children will persist in care who, due to circumstances and behavior,
are less likely to exit. Accordingly, the pattern of exits will decline
over time.
However, a child's time in foster care is time limited, usually not
extending beyond age 18. It is not unusual for older foster children to
receive emancipation and move to independent living arrangements
outside foster care. These considerations suggest that in time the exit
curve will begin to turn up; thus the underlying exit curve for foster
children will be an inverted U shape. We attempt to capture these
effects by introducing linear and quadratic terms for total duration
2213
times into our exit models. We anticipate that the linear term will carry
a negative sign, indicating an initial declining risk of exit, and that the
quadratic term will carry a positive sign, indicating an increasing rate.
We also anticipate some underlying duration effect for change; an
increasing proportion of older, problematic foster children may be
associated with increasing risks of change for these children. It is not
clear why this pattern may invert in time (as is the case for exits).
Accordingly, we attempt to control for this effect in our change models
by introducing a linear, but not a quadratic, term. We anticipate this
linear term will carry a positive sign, indicating an underlying pattern of
increasing risk of change.
3.3. Models and analysis
We tested the hypotheses using parametric models of exit and
placement change processes (Amburgey et al., 1993, p. 62; Carroll &
Hannan, 2000, pp. 371–374), which were defined operationally in
terms of the instantaneous rate:
h i
r jk ðt Þ ¼ lim qj ðt; t þ Δt Þ=Δt ;
where qj is the discrete probability of a child experiencing a type k
event (exit or placement change) between t and Δt, conditional on
the child being at risk for the event at time t. The rates of placement
change and of exit are assumed to be a loglinear function of the
covariates to preclude negative predicted outcomes (Blossfeld &
Rohwer, 2002), initially giving models of the general form:
r j kðt Þ ¼ exp½βX t ;
where Xt is a vector of time-varying variables, and β is a vector of
parameters indicating the effects of the variables on the rate of
occurrence.
Models of this form make two potentially problematic assumptions
relevant to testing our hypotheses, viz., the timing and occurrence of
placement changes for each child are independent of each other, and
potentially important but unmeasured covariates are not excluded from
the analysis. A moment's reflection suggests these assumptions probably
are not viable. For placement changes, the second and subsequent
placement changes likely are influenced by, and therefore are different
from, the initial and subsequent changes. For unmeasured covariates,
there may be factors intrinsic to individual children—aspects of behavior
or quality of foster home placements—for which we lack data but which
bear on placement change or exit. As recent methodological work shows
(Box-Steffensmeier, De Boef, & Zorn, 2007; Hougaard, 1991, 1999, 2000;
Kelly & Lim, 2000), not controlling for these possible effects could result
in misleading or spurious findings because of incorrect estimates of
standard errors, biases in the estimated effects of measured covariates, or
misleading estimates of duration dependency (Box-Steffensmeier &
Jones, 2004).
Accordingly, we fit our exponential models to the data as shared
frailty models. Such models address the problem of within-subject
correlation by introducing a random covariate or “error term” (Blossfeld
& Rohwer, 2002) into the equation. Some individuals are intrinsically
more prone to experience the event of interest than others, and
distribution of these individual-specific effects can be approximated by
assigning a specific distribution presumed to characterize the “error
term” (Box-Steffensmeier et al., 2007). We follow this convention here.
Accordingly the general form for our models is re-specified as:
r j kðt Þ ¼ exp½βX t þ Z i ω;
where ω corresponds to the random effect, defined by the gamma
distribution, the most frequently used distribution in applied research
(Box-Steffensmeier & Jones, 2004). It is presumed to capture a given
observation's unobserved propensity towards the event of interest. Zi
2214 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219

identifies observations that share a common frailty because they Table 1

pertain to the same child. We estimated the parameters of the models Descriptive statistics for all spells (N = 20,997), terminal spells (N = 3448) and gender
(N = 3448).
using maximum likelihood methods implemented in Stata (Cleves,
Gould, & Gutierrez, 2002, pp. 292–298; Gutierrez, 2002; StataCorp, Variables Mean s.d.
2003a,b). Unemployment 6.633 1.946
Labor force participation 48.074 7.369
4. Results Institutional change 0.692 0.462
Economic incentives 0.921 0.27
Labor unrest 0.01 0.098
Table 1 presents selected descriptive statistics on the study Matching ratio 2.253 0.151
population, including means and standard deviations for all spells Age clock 8.32 0.464
and terminal spells, as well as percent distribution for gender (53.34% Change 0.756 0.429
male). Table 2 gives an all-spells intercorrelation matrix of indepen- Cumulative change 5.851 9.464
Change clock 2.591 2.855
dent and control variables. It suggests that although the majority of
Age × Change 6.414 3.565
the correlation coefficients are positive and significant, multicolli- Terminal spells (n = 3448)
nearity does not appear to be a problem. Age in days (logged) 8.248 0.566
Table 3 presents information on the pattern of placement changes. Cumulative change 3.295 6.719
It shows that the majority of children, approximately 44%, experi-
Gender Male Female
enced only one change during their tenure in foster care. Approxi- 53.34 46.66
mately 79% had five or fewer changes. Of the remaining 720 children,
42.6%, or 307 children, experienced more than 10 placement changes,
with 112 of these having more than 20 changes. change, followed by a more rapid decline than would be expected
Table 4 displays estimated coefficients for exit rate models and because of aging. While these changes show a pattern of repetitive
information relevant to assessing heterogeneity in the data. Model I is momentum slowed by endogenous time decay in risk of exit, they
the baseline model. It gives the coefficients for gender, the control also show that effects of increasing age amplifies the overall pattern
variables, the dummy variable for whether or not there has been a of change. Considered together, therefore, these findings support
placement change, and the age clock. Model II adds elapsed time from Hypotheses 5 and 6, which proposed, respectively, that probability of
last change (change clock), plus the interaction term for placement placement change would decrease with time since the last change,
change and age. and that probability of placement change would increase with age of
Starting with Hypothesis 1 and the question of how placement child.
change affects exits rates, estimates in model I indicate that, as predicted Finally, the interaction term for change and age is positive and
by Hypothesis 1, children with at least one placement change had lower significant, indicating that early placement change increases likelihood
risk of exit compared to those with no changes. In model II, the of another change, thereby increasing momentum for further change.
interaction term for change and age is positive and significant, indicating This supports the argument for Hypothesis 7, which predicted that
that as a child ages into foster care, change is likely to be associated early placement change increases prospects for subsequent place-
within increasing risk of exit. This contradicts Hypothesis 2a, which ment changes.
anticipated that increasing age would act to lower the exit rate, but Similar to the exit models I and II in Table 4, the likelihood ratio test
supports Hypothesis 2b, which made the opposite prediction. Finally, the of theta= 0 for the change model is strongly statistically significant,
coefficient for the change clock in model II is positive and significant, indicating unobserved heterogeneity in the data. This finding also
indicating that, after a change, the risk of exit increases over time. Thus, supports that the effects associated with the measured covariates in
as anticipated by Hypothesis 3, the longer the time since the last change, Table 5 are not spurious. As a result, estimates reported in this table can
the higher the risk of exit. be construed as providing strong support for Hypotheses 4, 5, 6, and 7.
Regarding heterogeneity, the likelihood ratio test of theta= 0 is
strongly statistically significant for both models, indicating that consid- 5. Discussion
erable unobserved heterogeneity exists in the data, but this heteroge-
neity is not captured by the covariates in the models. The presence of We began by proposing to test AT as transactional and process
statistically significant effects for the relevant covariates described in the theory of change, particularized to foster children and to the way
preceding paragraph also tells us that controlling for this heterogeneity change occurs in their lives. In constructing this test, we presumed
does not eliminate our findings. As Blossfeld and Rohwer (2002, p. 269) that the effects of change processes in foster care work independently
point out: “The utility of estimating a mixture (frailty) model lies not in from the effects of child and other contextual attributes. Our analysis
separating an error term from a ‘true’ transition rate, but in providing strongly supports the efficacy of treating AT, in the context of foster
some means for assessing the robustness of estimates for the observed care, as a transactional process theory of change. For all hypotheses
covariates.” Accepting this argument implies that, overall, findings except Hypothesis 2a, the predicted effects of change were in the
presented in Table 4 strongly support Hypotheses 1, 2a, and 3, but not expected direction.
Hypothesis 2a. Hypothesis 2a predicted that older age at time of placement
Table 5 displays the results for the rates-of-change analysis. The change would be associated with a decreasing risk of exit. The
coefficient for cumulative placement changes is positive and signif- theoretical basis for this hypothesis was a fairly literal interpretation
icant, indicating that one placement change increases likelihood of of Bowlby's assertion of a positive relationship between children's
another placement change, and that effects are cumulative. This increasing age and increasingly inflexible behavior patterns. The
supports Hypothesis 4, which predicted that probability of placement competing hypothesis, Hypothesis 2b—based on Crittenden's dy-
change increases with number of previous placement changes. namic maturational model of attachment—predicted an increasing
Second, the change clock coefficient is significant and negative; risk of exit, and was supported by the analysis. Our finding of support
the main effect of age is positive, indicating that increasing propensity for Hypothesis 2b, and thus Crittenden's argument, is important for
for placement change as children age. These findings indicate that the two reasons. First, Crittenden does not dispute Bowlby's central claim
rate of these changes shows repetitive momentum, and that aging that children seek relational security in their lives, or the assumption
augments this pattern. The significant negative coefficient for the that the basis and orientation of a child's current behavior is strongly
change clock indicates an initial jump in the likelihood of another connected to the basis and orientation of earlier behavior. Hence, it does
 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219 2215

Table 2
Intercorrelation matrix all observations (N = 20,997).

Variables 1 2 3 4 5 6 7 8 9 10 11 12 13

1. Gender
2. Unemployment − 0.0186⁎
3. Labor force − 0.0325⁎ 0.1381⁎
participation
4. Institutional change − 0.0273⁎ 0.4104⁎ 0.7960⁎
5. Economic incentives − 0.0217⁎ 0.2860⁎ 0.4125⁎ 0.4400⁎
6. Labor unrest − 0.0026 0.1491⁎ 0.0408⁎ 0.0662⁎ 0.0291⁎
7. Matching ratio − 0.0244⁎ 0.1617⁎ 0.7010⁎ 0.6561⁎ 0.4687⁎ 0.0755⁎
8. Age clock 0.1515⁎ 0.1014⁎ 0.2095⁎ 0.2178⁎ 0.1668⁎ − 0.0788⁎ 0.1621⁎
9. Duration 0.1284⁎ 0.0948⁎ 0.2147⁎ 0.2115⁎ 0.1296⁎ − 0.0476⁎ 0.1645⁎ 0.9344⁎
10. Duration squared − 0.0865⁎ 0.0181⁎ 0.1957⁎ 0.1457⁎ 0.1138⁎ − 0.0322⁎ 0.1437⁎ 0.2268⁎ 0.2088⁎
11. Change 0.0062 0.0836⁎ 0.0976⁎ 0.0997⁎ 0.0632⁎ 0.0033 0.0828⁎ 0.0360⁎ 0.0108 0.0284⁎
12. Cumulative change − 0.0633⁎ 0.0196⁎ 0.3671⁎ 0.2761⁎ 0.1540⁎ − 0.0614⁎ 0.2698⁎ 0.2714⁎ 0.2288⁎ 0.3133⁎ 0.1580⁎
13. Change clock 0.0937⁎ 0.0476⁎ − 0.0809⁎ − 0.0324⁎ 0.0469⁎ − 0.0756⁎ − 0.0449⁎ 0.4010⁎ 0.2609⁎ − 0.0301⁎ − 0.0461⁎ − 0.1735⁎
14. Age × Change − 0.0447⁎ 0.0876⁎ 0.2900⁎ 0.2488⁎ 0.2233⁎ − 0.1531⁎ 0.2380⁎ 0.3756⁎ 0.2884⁎ 0.6093⁎ 0.1976⁎ 0.4451⁎ 0.0583⁎
⁎ p b .05.

not contradict our argument that children tend toward relative is amplified by the effects of increasing age, and the main effect of age
behavioral inertia. Second, Crittenden supports the idea that, in dealing significantly increases likelihood of change.
with variation over time in their felt-sense of security, children evolve a Hypotheses 7, which predicted that early change would increase
broader range of attachment strategies than anticipated by the fourfold probability of subsequent changes, was also supported by our analysis.
classification system used to characterize the attachment strategies of An early change added an increment to the likelihood of further changes,
very young children (Thompson & Raikes, 2003). Thus, Crittenden's and this increment increased with age of child. The Bowlby/Sroufe
argument affirms our emphasis on conceptualizing AT as a lifespan pathways model, which provides the theoretical basis for this hypothesis,
theory of change with potential to deepen understanding of children's
behavior at least to young adulthood. Table 4
Hypotheses 1 and 3 also support the effects of change on propensity Effects of placement change on exit●.
to exit. For Hypothesis 1, the disruptive effects of change for a child's Variable Model
internal working model increase the prospects for that child being
I II
labeled problematic, thus decreasing risk of exit. In contrast, Hypothesis
3 finds that, providing that the child stays in the same placement, the Gender − 0.181⁎⁎⁎ − 0.181⁎⁎⁎
(0.060) (0.063)
disruptive effects of change attenuate over time and increase chances of
Unemployment rate 0.140⁎⁎⁎ 0.138⁎⁎⁎
exit. (0.016) (0.017)
This latter finding raises a troubling issue. Specifically, a placement Women in labor force 0.010 0.013⁎
persists after a change because the child and the foster family have (0.007) (0.007)
achieved some level of mutual accommodation, resulting in the child Legal change 0.072 0.111
(0.100) (0.103)
being labeled non-problematic. Thus, conditions are created that support Economic incentives 0.673⁎⁎⁎ 0.683⁎⁎⁎
the child leaving foster care and returning to his or her own home. (0.099) (0.103)
Current policies and practices define this kind of effect as normatively Labor unrest − 0.932⁎⁎⁎ − 0.821⁎⁎⁎
appropriate. Reflection on the theory and findings reported in this paper, (0.265) (0.280)
Matching ratio − 0.022 − 0.144
however, suggests that it is not transparently clear that this is always in
(0.229) (0.237)
the child's best interests. We explore this issue in greater detail in the last Time − 2.245⁎⁎⁎ − 2.914⁎⁎⁎
paragraph of this discussion section. (0.086) (0.116)
Hypotheses 4, 5 and 6 support a “liability of change” argument. That Time squared 0.458⁎⁎⁎ 0.596⁎⁎⁎
is, change tends to engender change, and children tend to move from one (0.024) (0.029)
Age clock 0.007⁎⁎⁎ 0.004⁎⁎
context to another once change has been initiated. In addition, a process (0.002) (0.002)
of repetitive momentum operates for placement changes; more frequent Placement change − 5.086⁎⁎⁎ − 5.366⁎⁎⁎
experience with change increases the probability of its recurrence. At the (0.106) (0.112)
same time, this repetitive momentum is dampened somewhat between Change clock 0.302⁎⁎⁎
(0.028)
changes by a propensity for risk of change to decline with the passage of
Change × Age 0.005⁎⁎⁎
time. This overall pattern of change can be visualized as a series of jumps (0.002)
with a declining propensity to change following each jump. The pattern Constant − 5.508⁎⁎⁎ − 5.240⁎⁎⁎
(0.510) (0.535)
Theta 0.878⁎⁎ 1.056
Table 3 (0.047) (0.056)
Total number of placement experiences by children in care (N = 3448). Chi-squared 8057.62 8185.41
Degrees of freedom 11 13
Total number of placements Number of children % of Children Cumulative %
Number of children 3448 3448
1 1525 44.23 44.23 Log likelihood −3759.50 − 3695.60
2 429 12.44 56.67 Number of events 3170 3170
3 372 10.79 67.46 Number of observations 20,997 20,997
4 237 6.87 74.33 Likelihood-ratio test of theta = 0 733.22 712.46
5 165 4.79 79.12
⁎ p b .10.
6–10 413 11.98 91.1
⁎⁎ p b .05.
11–20 195 5.66 96.75
⁎⁎⁎ p b .01.
> 20 112 3.25 100 ●
Standard errors are in parentheses.
2216 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219

Table 5 more in relational terms (Sameroff & Emde, 1989); and the efficacy
Effects of change and time since change on rates of change●. of taking explicit account of attachment histories when assessing
Variable Model problems and developing both preventive programs and treatment
plans.
Gender 0.012
(0.044) Despite consistent support for our hypotheses, the analysis nonethe-
Unemployment rate 0.019⁎⁎⁎ less results in some qualification to our argument of behavioral inertia—
(0.007) at least as this inertia overtly manifests. Specifically, the findings show
Women in labor force − 0.053⁎⁎⁎
that risks of exit and change both depend on age; age moderates
(0.004)
Legal change − 0.191⁎⁎⁎ between change and exit, and between earlier and later change. In the
(0.052) case of exits, the main effect of age is positive; for change, it is negative.
Economic incentives − 0.182⁎⁎⁎ As a consequence, effects of age are more complex than anticipated.
(0.060) Older children are more likely to exit and, therefore, have a higher base
Labor unrest − 0.180⁎
hazard rate. As a result, the net effect of change is likely to be quite
(0.111)
Matching ratio − 0.284⁎⁎ powerful. It produces a more marked change in an older child's risk of
(0.113) exiting compared to the effects of similar change in a younger child,
Time 0.618⁎⁎⁎ because the latter has a lower base hazard rate. On the other hand,
(0.022)
when it comes to change, the situation is reversed. Older children are
Age clock − 0.040⁎⁎⁎
(0.002) less likely to experience change since they have a lower base hazard
Cumulative placement changes 0.008⁎⁎⁎ rate. Accordingly, the net effect on rates of change is likely to be stronger
(0.001) among younger children, who start from a higher base hazard rate.
Change clock − 0.345⁎⁎⁎ One plausible interpretation of this pattern, in line with the earlier
(0.011)
discussion of Crittenden's dynamic maturational model of attachment, is
Change × Age 0.004⁎⁎⁎
(0.001) that increasing age brings increased sophistication by children adapting
Constant 9.595⁎⁎⁎ to and/or influencing their context. They may use this to negotiate exit
(0.268) from foster care at opportune times, or to facilitate heightened stability
Theta 0.758⁎⁎⁎
of placement when they persist in care.
(0.043)
Chi-squared 3723.19
In a related vein, the significance of elapsed time between successive
Degrees of freedom 12 changes also merits consideration. When the change clock is reset by a
Number of children 3448 change, the elapsed time since the last change is a stronger indicator of
Log likelihood 897.09 the cumulative effects of being in care than age is. Accordingly, change
Number of events 11,443
has long-term effects on the hazard rate on which age-associated effects
Number of observations 20,997
Likelihood-ratio test of theta = 0 2264.42 are based. This suggests that underlying processes associated with
children's placement patterns and disposition are not simply condi-
⁎ p b .10.
⁎⁎ p b .05. tioned by chronological aging; they also strongly reflect the influence
⁎⁎⁎ p b .01. that change experiences have on children.
●
Standard errors are in parentheses. Interestingly, this pattern of change and exit, and the potential
significance of relational experience in explaining it, is in line with
arguments and research concerning self-fulfilling prophecies in close
relations (Downey, Freitas, Michaelis, & Khouri, 1998; Sroufe, 1990). In
has helped support and rationalize a widely accepted approach to the brief, arguments assert that, conditioned by their relational experiences,
study of childhood problems. This approach focuses on identifying and people's internal working models of relationships incorporate expecta-
understanding factors and processes that, over time, augment a child's tions of acceptance and rejection, which then shape their relationships in
chances of moving to maladaptive pathways (leading to behavioral the manner of a self-fulfilling prophecy. Sensitivity to expectations of
disturbances) or deflecting them away from such pathways (Schofield rejection by others leads people to behave in ways that solicit such
& Beek, 2005). rejection. This affirms and perpetuates a disposition to expect, perceive,
Significantly, however, we are not aware that the underlying premise and overreact to rejection. Conversely, expectations of acceptance lead to
of this model—an initial deviant pathway change leads to successive behaviors more solicitous of being accepted by others, and less to a
deviant changes that take children away from pathways leading to disposition to behave as if expecting rejection.
competent social functioning—has been tested empirically. To the extent Applying this self-fulfilling argument to AT as a process theory of
that early placement change in foster care can be construed as reflecting change, it is plausible that, for many children, being taken from their
maladaptive behavior patterns (we would argue they can), our finding families into foster care is a powerful rejection experience. It may
for Hypothesis 7 may be the first empirical test of the validity of this compound intra-familial rejection experiences that made child
underlying premise. welfare authority action necessary in the first place. Hence, if it is
Accepting ours as the first affirming empirical test suggests at not already present, it is probable that many foster children start
least two things. First, some additional confidence can be placed in foster care tenure with a heightened sensitivity to the possibility of
earlier and future research findings testing empirical implications rejection. Subsequent placement changes, particularly if place-
based on the pathways developmental model. Process, it seems, does ments are short, likely will be perceived as affirming the necessity
count, making it possible, at least in the case of foster children, to for rejection sensitivity, thereby orienting internal working models
predict outcomes based on knowledge of their context and patterns of relationships to defensively expect rejection from others. This
of change without explicit and detailed knowledge of their personal augments the probability of confirmatory behavior from others, and
attributes. Second, it adds significance to Sroufe's (1997) arguments conditions appropriate for the emergence of the self-fulfilling
regarding the potential importance of adopting a developmental prophecy described above are put in place.
approach to psychopathology in children instead of relying almost Empirically this would manifest in the pattern of findings reported
exclusively on the dominant medical model. Among other things, in this research. Based on the arguments presented above, assuming
this suggests need for expanded study of continuity and change in that children in foster care vary in their levels of rejection sensitivity
children's lives; the relevance of conceptualizing their problems implies that those with lower levels will be less likely to perceive
 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
placement change defensively. Thus, in the longer run, they will be
more likely to solicit acceptance behaviors from caregivers, leading to
placement stability. This will be reflected empirically in longer times
in placements and lower rates of change.
Conversely, those with higher levels of rejection sensitivity will be
more prone to perceive change defensively and overreact to it, thereby
prompting confirmatory responses. Empirically, this would manifest in
progressively shorter times in placements and an increasing propensity
for one change to be followed by another.
Though we cannot test the validity of this explanation here, it does
suggest an important approach for future research that—building on the
contributions of this and other research—could illuminate the mecha-
nisms and processes underlying how the experience of change influences
children's life trajectories and outcomes. To ensure a strong test of
the theory, such research should not only account for rejection and
acceptance expectancies of children, but the rejection and accep-
tance expectancies of caregivers and professionals who work with
the children and are involved in planning changes in their lives.
Children come with histories. Considered in the context of the arguments
and findings presented in this research, how children's significant others
interpret those histories may be a crucial element in determining the
nature and direction of the developmental pathways that subsequently
evolve.
Returning now to the promised discussion of Hypothesis 3, the issue
concerned implications of the finding that increased stability in foster
care increased the likelihood of the child exiting foster care. Are such
transitions in the child's best interests? Normatively current policies and
practices give an affirmative answer. We think the theory and findings of
this research suggest caution in applying this view. Specifically, stability
in a placement indicates the presence of a relationship between the
foster child and caregivers. Moving the child out of this environment
disrupts that relationship, regardless of the purpose of the move. In
addition to disrupting a relationship, it also transitions the child to a
context that, while entailing a historical connection, may now appear as
new or strange, requiring the renegotiation of old relationships or even
the establishment of new ones.
Our theory suggests that this combination of disrupting old
relationships and the renegotiating or creating new ones is likely to
induce anxiety, potentially leading to child behaviors inimical to
persistence in this new environment, with a subsequent return to
foster care. Considered in the context of findings and discussion
presented in this study, the implications are that, all other things
assumed equal, such transitions carry the potential to do more harm
than good. Because of the strong preference afforded this approach,
transitions that move children from stable foster home environ-
ments to their home environments seem a significant topic for future
research.
6. Contribution, limitations, and conclusions
The results of this study are important for several reasons. They
affirm the importance and relevance of studying effects of change
processes, not just the content of change, in children's lives. While
characteristics of children and their contexts did have effects, the
consequences of changes examined did not depend only on these
characteristics and how they changed over time.
Significantly, the effects produced by change operated indepen-
dently of the effects of these characteristics during the time children
were in care. In addition, results suggest that a strong historical
dependency exists in the trajectory of children's lives. This finding
clearly aligns with arguments that history matters in accounting for
current child behavior. From the perspective of this research, future
research that omits information about the history of significant
changes in children's lives, and how such changes occur, will be
limited in its capacity to explain their current circumstances.
2217
Another refinement to future foster care research, consistent with
this paper's view of AT as a life span theory, is suggested by recent
theoretical and philosophical analyses of temporality in social research
(Abbott, 2001; Clemens, 1999). It supports conceptualizing a child's life
trajectory as a sequence of nested durations marked by events that
reach backward and forward in their causes and effects (Abbott, 2001:
226). From this perspective, and for example, adoption and reunifica-
tion are not outcomes or “ends” in a foster child's life trajectory. Instead
they are “middles” with something before and something after (Abbott,
2001, 2005; MacKenzie & McDonough, 2009). This dialectical issue is
obscured by the periodization of foster care to durations from entry to
event-specific exits such as adoption or reunification conceptualized as
outcomes of the foster care system. Hence, future research into the
dynamics of foster care should expand its scope to include questions
about how events such as adoption and reunification affect a child's
subsequent development and life chances.
While the results of this study do contribute to attachment and
change-and-stability literature, certain cautions are in order. In modeling
our data, we were constrained to assume that the error term in our eqs.
could be accurately described using a gamma distribution. Because
introducing a different distribution may have given different results,
there are limitations to the authoritativeness of our findings. In addition,
our emphasis has been on exploring AT usefulness in deepening
understanding of special populations of children, in this case, foster
children. Hence, while we would construe our findings as applicable
to foster children generally, we can make no such claim in the case of
other populations of children.
We are limited in drawing theoretical inferences about the meaning
of our findings. The results of this study show that a child's age is an
important consideration in explaining the frequency and magnitude of
change. However, a substantial literature suggests that aging in children
is more than a chronological process specific to the individual (Dannefer,
1984; Featherman & Lerner, 1985; Laz, 1998; Neugarten, Moore, & Lowe,
1965; Riley, 1987). It is social, contextual, and has strong potential to
influence children's developmental pathways. We lacked the informa-
tion needed to disentangle these effects, suggesting a less precise
analysis of their comparative influences than perhaps should be the case
(Riley, 1987). In a related vein, we also lacked direct measures of
schemas and, therefore, of possible changes to internal working models
as a result of changes in schema organization patterns. This persistent
problem in attachment research is worth noting. Without its resolution,
we continue to struggle to differentiate empirically between internal
and external effects and consequences in studies that accord central
explanatory roles to internal working models.
We began this paper with two critical claims. First, modeling
variation over time in foster child exit and placement change rates, as
explained by such demographic attributes as age, was complicated
by the fact the same kinds of connections also could explain the
distribution of such attributes in the study population. Second, AT
was a relevant framework for explicating and understanding the
dynamic, reciprocal processes implied as underlying patterns of
foster child exit and change. The results of this research strongly
affirm the latter observation. By giving insight into children's
developmental processes, AT enabled us to show, for our study
population, that how children are affected by change is crucially
determined by when it happens. Moreover, by giving a basis for
deriving this answer, AT also provides practical insight for those
who work with foster children. Like history, change counts. And,
for this vulnerable group, its effects should never be treated as
benign.
References
Abbott, A. (2001). Time matters. Chicago: University of Chicago Press.
Abbott, A. (2005). Process and temporality in sociology: The idea of outcome in US
sociology. In G. Steinmetz (Ed.), The politics of method in the human sciences
(pp. 392–426). Durham: Duke University Press.
2218 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
Akin, B. A. (2011). Predictors of foster care exits to permanency: A competing risks
analysis of reunification, guardianship, and adoption. Children and Youth Services
Review, 33(6), 999–1011.
Amburgey, T. L., Kelly, D., & Barnett, W. P. (1993). Resetting the clock: The
dynamics of organizational change and failure. Administrative Science Quarterly,
38(1), 51–73.
Blossfeld, H. P., & Rohwer, G. (Eds.). (2002). Techniques of event history modeling: New
approaches to causal analysis (2nd ed.). Lawrence Erlbaum AssociatesMahwah, NJ.
Bowlby, J. (1969). Attachment. New York: Basic Books.
Bowlby, J. (1973). Attachment and loss: Separation (2nd ed.). New York: Basic Books.
Bowlby, J. (1980). Attachment and loss: Loss. New York: Basic Books.
Bowlby, J. (1988). A secure base: Parent–child attachment and healthy human
development. New York: Basic Books.
Box-Steffensmeier, J. M., De Boef, S., & Zorn, C. J. W. (2007). Event dependence and
heterogeneity in duration models: The conditional fraility model. Political Analysis,
15(3), 237–256.
Box-Steffensmeier, J. M., & Jones, B. S. (2004). Event history modeling: A guide for social
scientists. New York: Cambridge University Press.
Bretherton, I., & Munholland, K. A. (1999). Internal working models in attachment
relationships: A construct revisited. In J. Cassidy, & P. R. Saver (Eds.), Handbook of
attachment: Theory, research and clinical applications (pp. 89–111). New York: The
Guilford Press.
Bretherton, I., & Waters, E. (1985). Attachment theory: Retrospect and prospect.
Monographs of the Society for Research in Child Development, 50(1–2), 3–35.
Carroll, G., & Hannan, M. T. (2000). The demography of corporations and industries.
Princeton, N.J.: Princeton University Press.
Child Welfare Information Gateway (2010). Foster Care Statistics 2009. Retrieved
February 2, 2011, from. U.S. Department of Health & Human Services, Adminis-
tration for Children and Families http://www.childwelfare.gov
Children's Bureau (2010). Child welfare outcomes 2004–2007: Report to Congress.
Washington, DC: U.S. Department of Health & Human Services.
Cicchetti, D., Rogosch, F. A., Maughan, A., Toth, S. L., & Bruce, J. (2003). False belief
understanding in maltreated children. Development and Psychopathology, 15(4),
1067–1091.
Clemens, E. S. (1999). Continuity and coherence: Periodization and the problem of
institutional change. In F. Engelstad, & R. Kalleberg (Eds.), Social time and social
change: perspectives on sociology and history (pp. 62–83). Oslo: Scandinavian
University Press.
Cleves, M. A., Gould, W. W., & Gutierrez, R. G. (2002). An introduction to survival analysis
using STATA. College Station, TX: Stata Corporation.
Connell, C. M., Vanderploeg, J. J., Flaspohler, P., Katz, K. H., Saunders, L., & Kraemer
Tebes, J. (2006). Changes in placement among children in foster care: A
longitudinal study of child and case influences. The Social Service Review, 80(3),
398–418.
Crick, N. R., & Dodge, K. A. (1994). A review and reformulation of social
information-processing mechanisms in children's social-adjustment. Psychological
Bulletin, 115(1), 74–101.
Crittenden, P. M. (1999). Danger and development: The organization of self-protective
strategies. Monographs of the Society for Research in Child Development, 64, 145–171
(Chap. 3).
Crittenden, P. M. (2000). Dynamic-maturational exploration of the meaning of security
and adaptation: Empirical, cultural and theoretical considerations. In P. M.
Crittenden, & A. H. Claussen (Eds.), The organization of attachment relationships
(pp. 358–383). Cambridge, UK: Cambridge University Press.
D'Andrade, A. C. (2005). Placement stability in foster care. In G. P. Mallon, & P. M. Hess
(Eds.), Child welfare for the 21st century: A handbook of practices, policies, and
programs (pp. 608–622). New York: Columbia University Press.
Dannefer, D. (1984). The role of the social in life-span developmental-psychology, past
and future: Rejoinder. American Sociological Review, 49(6), 847–850.
Dodge, K. A., Pettit, G. S., Mcclaskey, C. L., & Brown, M. M. (1986). Social competence in
children. Monographs of the Society for Research in Child Development, 51(2), 1–85.
Downey, G., Freitas, A. L., Michaelis, B., & Khouri, H. (1998). The self-fulfilling prophecy
in close relationships: Rejection sensitivity and rejection by romantic partners.
Journal of Personality and Social Psychology, 75(2), 545–560.
Dozier, M., Albus, K., Fisher, P. A., & Sepulveda, S. (2002). Interventions for foster
parents: Implications for developmental theory. Development and Psychopathology,
14(4), 843–860.
Featherman, D. L., & Lerner, R. M. (1985). Ontogenesis and sociogenesis: Problematics
for theory and research about development and socialization across the lifespan.
American Sociological Review, 50(5), 659–676.
Fichman, M., & Levinthal, D. A. (1991). Honeymoons and the liability of adolescence: A
new perspective on duration dependence in social and organizational relation-
ships. Academy of Management Review, 16, 442–468.
Gutierrez, R. G. (2002). Parametric frailty and shared frailty survival models. The Stata
Journal, 2(1), 22–44.
Horwitz, S. M., Balestracci, K. M. B., & Simms, M. D. (2001). Foster care placement
improves children's functioning. Archives of Pediatrics & Adolescent Medicine,
155(11), 1255–1260.
Hougaard, P. (1991). Modeling heterogeneity in survival-data. Journal of Applied
Probability, 28(3), 695–701.
Hougaard, P. (1999). Fundamentals of survival data. Biometrics, 55(1), 13–22.
Hougaard, P. (2000). Analysis of multivariate survival data: Statistics for biology and
health. New York: Springer.
Howes, C. (1999). Attachment relationships in the context of multiple caregivers. In J.
Cassidy, & P. R. Shaver (Eds.), Handbook of attachment theory, research, and clinical
applications (pp. 671–687). New York: Guilford Press.
Howes, C., & Segal, J. (1993). Children's relationships with alternative caregivers: The
special case of maltreated children removed from their homes. Journal of Applied
Developmental Psychology, 14, 71–81.
James, S. (2004). Why do foster care placements disrupt? An investigation of reasons
for placement change in foster care. The Social Service Review, 78(4), 601–627.
Jovanovic, B. (1979). Job matching and the theory of turnover. Journal of Political
Economy, 87, 972–990.
Kelly, P. J., & Lim, L. L. Y. (2000). Survival analysis for recurrent event data: An
application to childhood infectious diseases. Statistics in Medicine, 19(1), 13–33.
Land, K. C. (1984). Markovian transition rates models of macro social change. In A.
Diekmann, & P. Mitter (Eds.), Stochastic modeling of social processes (pp. 215–243).
Orlando, FL: Academic Press.
Laudan, L. (1977). Progress and its problems: Towards a theory of scientific growth.
Berkeley: University of California Press.
Laz, C. (1998). Act your age. Sociological Forum, 13(1), 85–113.
Levitt, B., & March, J. G. (1988). Organizational learning. Annual Review of Sociology, 14,
319–340.
Lewis, M. (2001). Issues in the study of personality development. Psychological Inquiry,
12(2), 67–83.
Lewis, M., Feiring, C., & Rosenthal, S. (2000). Attachment over time. Child Development,
71(3), 707–720.
Lyon, T. D. (1987). Angelino foster care children's understanding of their removal from
home. Unpublished Report. Southern California Studies Center, University of
Southern California.
MacKenzie, M. J., & McDonough, S. C. (2009). Transactions between perception and
reality: Maternal beliefs and infant regulatory behavior. In A. J. Sameroff (Ed.), The
Transactional Model of Development: How children and contexts shape each other
(pp. 35–54). Washington: APA Books.
Macrae, C. N., & Bodenhausen, G. V. (2000). Social cognition: Thinking categorically
about others. Annual Review of Psychology, 51, 93–120.
Main, M., Kaplan, N., & Cassidy, J. (1985). Security in infancy, childhood and adulthood:
A move to the level of representation. Monographs of the Society for Research in
Child Development, 50(1–2), 66–104.
March, J. C., & March, J. G. (1977). Almost random careers: Wisconsin school
superintendency, 1940–1972. Administrative Science Quarterly, 22(3), 377–409.
Markovits, H., Benenson, J., & Dolenszky, E. (2001). Evidence that children and
adolescents have internal models of peer interactions that are gender differenti-
ated. Child Development, 72(3), 879–886.
Marvin, R. S., & Britner, P. A. (1999). Normative development: The ontogeny of
attachment. In J. Cassidy, & P. R. Saver (Eds.), Handbook of attachment: Theory,
research and clinical applications (pp. 44–67). New York: The Guilford Press.
Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial-behavior: A
developmental taxonomy. Psychological Review, 100(4), 674–701.
Mrazek, P. J., & Mrazek, D. A. (1987). Resilience in child maltreatment victims: A
conceptual exploration. Child Abuse & Neglect, 11(3), 357–366.
Neugarten, B. L., Moore, J. W., & Lowe, J. C. (1965). Age norms, age constraints, and
adult socialization. The American Journal of Sociology, 70(6), 710–717.
Newton, R. R., Litrownik, A. J., & Landsverk, J. A. (2000). Children and youth in foster
care: Disentangling the relationship between problem behaviors and number of
placements. Child Abuse & Neglect, 24(10), 1363–1374.
Palmer, S. E. (1996). Placement stability and inclusive practice in foster care: An
empirical study. Children and Youth Services Review, 18(7), 589–601.
Pears, K. C., & Fisher, P. A. (2005). Emotion understanding and theory of mind among
maltreated children in foster care: Evidence of deficits. Development and
Psychopathology, 17(1), 47–65.
Pecora, P. J., Whittaker, J. K., Maluccio, A. N., Barth, R. P., & DePanfilis, D. (2009). The
child welfare challenge: Policy, practice, and research. New Brunswick, N. J: Aldine
Transaction Publishers.
Rhodes, K. W., Orme, J. G., & Buehler, C. (2001). A comparison of family foster parents
who quit, consider quitting, and plan to continue fostering. The Social Service
Review, 75(1), 84–114.
Riley, M. W. (1987). On the significance of age in sociology. American Sociological
Review, 52(1), 1–14.
Rutter, M., & O'Connor, T. G. (1999). Implications of attachment theory for child care
policies. In J. Cassidy, & P. R. Shaver (Eds.), Handbook of attachment theory, research,
and clinical applications (pp. 823–844). New York: Guilford Press.
Sameroff, A. J., & Emde, R. N. (Eds.). (1989). Relationship disturbances in early childhood:
A developmental approach. New York: Basic Books.
Sameroff, A. J., & MacKenzie, M. J. (2003). Research strategies for capturing
transactional models of development: The limits of the possible. Development
and Psychopathology, 15(3), 613–640.
Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals, and understanding an inquiry
into human knowledge structures. Hillsdale, NJ: L. Erlbaum Associates.
Schofield, G., & Beek, M. (2005). Risk and resilience in long-term foster-care. British
Journal of Social Work, 35(8), 1283–1301.
Sroufe, L. A. (1990). An organizational perspective on self. In D. Cicchetti, & M. Beeghly
(Eds.), The self in transition: Infancy to childhood (pp. 281–307). Chicago: University
of Chicago Press.
Sroufe, L. A. (1997). Psychopathology as an outcome of development. Development and
Psychopathology, 9(2), 251–268.
Sroufe, L. A., Carlson, E. A., Levy, A. K., & Egeland, B. (1999). Implications of attachment
theory for developmental psychopathology. Development and Psychopathology,
11(1), 1–13.
Sroufe, L. A., Egeland, B., & Kreutzer, T. (1990). The fate of early experience following
developmental-change: Longitudinal approaches to individual adaptation in
childhood. Child Development, 61(5), 1363–1373.
 D.J. Tucker, M.J. MacKenzie / Children and Youth Services Review 34 (2012) 2208–2219
Sroufe, L. A., & Fleeson, J. (1986). Attachment and the construction of relationships. In
W. W. Hartup, & Z. Rubin (Eds.), Relationships and development (pp. 51–71).
Hillsdale, N.J: L. Erlbaum Associates.
StataCorp (2003a). Survival analysis and epidemiological tables reference manual:
Release 8. College Station, TX: Stata Corporation.
StataCorp (2003b). Stata statistical software: Release 8. College Station, TX: StataCorp
LP.
Steele, B. F. (1986). Notes on the lasting effects of early child-abuse throughout the
life-cycle. Child Abuse & Neglect, 10(3), 283–291.
Steinhauer, P. D. (1991). The least detrimental alternative: A systematic guide to case
planning and decision-making for children in care. Toronto: University of Toronto
Press.
Swan, C. A., & Sylvester, M. S. (2006). The foster care crisis: What caused caseloads to
grow? Demography, 43(2), 309–335.
Thompson, R. A., & Raikes, H. A. (2003). Toward the next quarter-century: Conceptual
and methodological challenges for attachment theory. Development and Psychopa-
thology, 15(3), 691–718.
Tucker, D. J., & Hurl, L. F. (1992). An ecological study of the dynamics of foster home
entries. The Social Service Review, 66(4), 617–641.
2219
Tucker, D. J., Hurl, L. F., & Ford, H. (1994). Applying organizational ecology to the
family: The case of who persists in providing foster-care. Journal of Marriage and
the Family, 56(4), 1005–1018.
van Santen, E. (2010). Predictors of exit type and length of stay in non-kinship family
foster care: The German experience. Children and Youth Services Review, 32(10),
1211–1222.
Verschueren, K., Marcoen, A., & Schoefs, V. (1996). The internal working model of the
self, attachment, and competence in five-year-olds. Child Development, 67(5),
2493–2511.
Vuchinich, S., Teachman, J., & Crosby, L. (1991). Families and hazard rates that change
over time: Some methodological issues in analyzing transitions. Journal of Marriage
and the Family, 53(4), 898–912.
Weinfield, N. S., Sroufe, L. A., & Egeland, B. (2000). Attachment from infancy to early
adulthood in a high-risk sample: Continuity, discontinuity, and their correlates.
Child Development, 71(3), 695–702.
Weinstein, E. A. (1960). The self-image of the foster child. New York: Russell Sage
Foundation.