Sibling Separation and Placement Instability for Children in Foster Care

Abstract

U.S. foster care policy prioritizes keeping siblings together while in foster care. However, prior research on the effects of sibling placement is limited in sample, measures, and research design. In this study, we use data on 2,297 children from an urban county in years 2015-2019 and assess how sibling separation is associated with placement instability. We use multilevel parametric hazard modeling with adjustments for child, sibling, and placement characteristics. Findings indicate that children placed with at least one sibling are less likely to experience a placement move and are specifically less likely to experience a non-progress move (e.g., moves due to problems or negative experiences in their foster home). For larger sibling groups, sibling separation was not consistently associated with placement instability and there was little difference in placement instability for children placed with some versus all siblings. Results were robust to differences in measurement and model specification. Black or Hispanic race/ethnicity was also associated with increased risk of instability, and associations between sibling separation and instability were stronger for Black children, implying enhanced efforts to maintain sibling groups may be especially beneficial for Black children. Overall, findings provide support for the continuation and expansion of policies promoting sibling placement.

For children in foster care, federal and state policy, as well as best practice guidelines, promote placement with siblings as a strategy for improving children’s experiences and outcomes in foster care. Even when necessary to protect child health and safety, foster care can be disruptive and distressful due to separation of children from their familiar physical environments and primary caregivers. Maintaining children’s other key connections, like their sibling relationships, may help children to cope with those losses and facilitate positive adjustment (Rock, Michelson, Thomson, & Day, 2015). However, to date, research on the associations between sibling placement or separation and children’s stability, permanency and wellbeing in foster care has been inconclusive, with significant limitations pertaining to sampling and methodology (Shlonsky, Bellamy, Elkins, & Ashare, 2005; DiGiovannia & Font, in press). This study assesses associations between sibling separation and placement instability for a cohort of children entering foster care between 2015 and 2019 in Allegheny County, Pennsylvania.

Sibling Separation in Foster Care

Siblings exert a significant influence on children’s functioning: children may look to older siblings as role models, and – in the case of neglectful or abusive family environments –siblings may be primary attachment relationships (McCormick, 2010; Rock et al., 2015). In recognition of these significant relationships, the federal Fostering Connections to Success and Increasing Adoptions Act (2008; P.L. 110-351) requires reasonable efforts to place siblings in the same foster, adoptive, or guardianship placement, and, where joint placement is not possible, to facilitate contact between separated siblings. The federal Act led at least 13 states to create new policies, in addition to at least 28 states that already had statutes requiring efforts to keep siblings together (Child Welfare Information Gateway, 2015).

A child in foster care may be separated from their siblings in one of three circumstances. First, separation may be intentional because it is contrary to the safety or wellbeing of one or more children to place them together. Second, reasonable efforts to place children together may be unsuccessful due to a lack of placement options. States face barriers to placing siblings together, particularly large sibling groups (U.S. Department of Health and Human Services [USDHHS], 2011) or sibling groups involving an older child or child with special needs (U.S. Government Accountability Office [USGAO], 2018). Third, siblings may be separated when only part of the sibling group is in foster care. This can occur when a child enters foster care for reasons related to their behavior (e.g., delinquency) or parent-child conflict, or when all siblings enter care together but do not exit at the same time (e.g., if children are pursuing different paths to permanency). When at least one but not all members of a sibling group are in foster care, the child in care does not have rights to sibling visitation (it is at the discretion of the person with legal custody of their siblings). To our knowledge, this form of separation has not been a focus of previous research.

Implications for Placement Instability

When children enter foster care, it can be difficult to adjust to a different environment – not only new caregivers and a new physical environment, but also new expectations, norms, and routines. Each placement change requires children to experience this adjustment period once more. Compounding these difficult adjustment processes, placement changes may evoke feelings of loss, rejection, and distrust, particularly when the placement change was unexpected or unplanned (Chambers et al., 2018). After multiple moves, children may also have more difficulty developing trust and connection with caregivers and act out more (Rubin, O’Reilly, Luan, & Localio, 2007), thus increasing their risk for additional moves (Aarons et al., 2010). Consequently, placement stability is a key priority in federal and state policy. The federal government tracks state performance on placement stability, using the proportion of children who experience two or fewer placements by the end of their foster care episode (USDHHS, 2011). However, given the limitations of relying on exit cohort data (Courtney, Needell, & Wulczyn, 2004), researchers have traditionally emphasized the hazard of, or time until, a placement move using survival analysis to enable inclusion of children who remain in care at the end of the observation period and account for different lengths of stay. More recently, studies have emphasized the distinction between moves that further policy or practice preferences for placement and moves that indicate a poorly-matched or inappropriate placement experience (Cross, Koh, Rolock, & Eblen-Manning, 2013; Font, 2015; James, 2004; Koh, Rolock, Cross, & Eblen-Manning, 2014). Following Font, Sattler, and Gershoff (2018), we refer to these as “progress moves” and “non-progress moves”. Progress moves include moves to live with kin, with siblings, or in a less-restrictive setting, whereas non-progress moves are attributed to an underlying problem with the original foster care placement, such as maltreatment in the placement, a child or caregiver requesting that the placement be changed, or a child requiring a more restrictive setting (e.g., residential facility).

Sibling separation may increase the probability of a progress move. Children initially unable to be placed with their siblings may be moved in order to be reunited with their siblings (a progress move), which would be a more mechanical effect of policies promoting sibling placement and would not per se provide any information about the extent to which sibling placement itself improves children’s experiences. Thus, our interest in this study is primarily with the question of whether, and to what extent, sibling placement reduces non-progress moves.

Scholars have argued that children may find it easier to adjust to a new environment when they have a familiar companion (McCormick, 2010); thus, the adjustment process may involve less acting out or fewer behavioral challenges – a key antecedent of placement moves (Konijn et al., 2019). When separated, children may feel angry or sad (Chambers et al., 2018) and therefore be more likely to have negative interactions within their foster home that result in placement changes. Potentially indicative of better adjustment for children placed with siblings, Sattler and colleagues (2018) found that older children were less likely to run away from placement or ask to be moved to a new home if their siblings were placed with them. Sibling placement also aligns with children’s preferences (Dickerson, Lyon, & Quas, 2019).

However, placement with siblings could exacerbate risk of non-progress moves. Although states are not required to pursue joint placement if contrary to the health or safety of a child, no federal guidelines address how to assess when situations warrant an exception (Administration for Children and Families, 2010). Evaluation of children’s best interests in placement decisions lacks transparency and consistent criteria (Font & Gershoff, 2020) and some siblings may be placed together in spite of harmful or abusive interactions. When this occurs, particularly without adequate supports to address health and safety concerns, the placement may disrupt due to risk of abuse or diminished functioning of one or more children. Second, joint placement of sibling groups may increase strain on foster families. Larger family size is a risk factor for abuse and neglect (Stith et al., 2009) and each additional child places a strain on families’ economic resources and confers greater demands on caregivers’ time and energy (Lawson & Mace, 2009). Because children in foster care often have special needs stemming from their prior exposures to abuse and neglect, such as developmental delays or behavioral challenges, the demands of an additional child may be especially high for foster families, resulting in increased stress and reduced quality of care. Consistent with this hypothesis, Sattler and colleagues (2018) found that children placed with siblings were more likely to move due to receiving substandard care than children without siblings or children placed apart from siblings. When siblings have different levels of behavioral or mental health challenges – which can occur when an older sibling experienced a significantly longer period of abuse or neglect than their younger sibling or when one sibling was the primary target for mistreatment – difficulty providing care for the higher-needs child may result in placement disruption for all siblings. In addition, given the limited supply of foster homes (USDHHS, 2011; USGAO, 2018), sibling placement may require trade-offs, such as changing schools or residing further away from birth parents (which complicates visitation arrangements), or placement with foster parents with less experience or training. Where such trade-offs exist, benefits of sibling placement may be offset by harms stemming from a poorly matched foster home.

Limitations of Prior Research and Contributions of the Present Study

A recent meta-analysis on predictors of placement instability concluded that sibling separation increases placement instability, based on nine quantitative studies published prior to 2015 (Konijn et al., 2019). However, research on sibling placement is difficult to synthesize due to variability in how sibling placement is measured, sample composition, length of observation, and other factors (Shlonsky et al., 2005). Specifically, many studies only include a ‘focal child’ rather than sibling groups (Hegar & Rosenthal, 2009, 2011) or focus on sibling pairs (Waid, Kothari, Bank, & McBeath, 2016). Others only have data on children in foster care, and thus cannot measure separation from siblings who remain at home (Font et al., 2018; Sattler et al., 2018). In measuring separation, some studies focus on placement alone versus placement with at least one sibling (Barth et al., 2007; Leathers, 2006); others measure separation from at least one sibling versus placement with all identified siblings (Sattler et al., 2018). A few studies used multiple categories to distinguish between full and partial separation or emphasized the consistency of placement together (Koh et al., 2014; Leathers, 2005). It is unclear whether inconsistent research findings reflect differences in study samples, methods, or measurement.

In addition, a recent scoping review of sibling placement research found that many studies of sibling placement and stability only reported bivariate correlations and those that included statistical controls did not account for characteristics of children’s siblings (e.g., age or disability status of sibling) that may affect both sibling separation and risk of instability (DiGiovanni & Font, in press). Lastly, few studies specifically modeled placement disruptions, or non-progress moves (versus any placement change; for exceptions, see: Font et al., 2018; Leathers, 2006; Sattler et al., 2018). Consideration of different types of moves is needed to discern whether differences in risk of placement changes for jointly placed siblings reflect inherent benefits of joint placement or efforts to reunite separated siblings. The present study addresses these limitations by (1) testing multiple measures of sibling separation; (2) using a complete cohort of children entering foster care with measures of both child and sibling group characteristics; and (3) measuring both ‘any placement move’ as is common in prior research, as well as non-progress moves, which are more likely indicative of adverse foster care experiences.

Method

Data and Sample

This study draws from a complete cohort of children reported to the child welfare system (CWS) between 2015 and 2019 in Allegheny County, Pennsylvania. Data were provided through a data use agreement with Allegheny County Department of Human Services and the study was approved by the Internal Review Board at the authors’ university. Children’s CWS involvement were tracked until the end of the observation period (12/31/2019). Individual children were identified largely by system-assigned IDs but additional information was used where IDs were inconsistent (e.g., one child had multiple IDs). Children were presumed to be siblings if they were ever listed on the same CWS case or referral as each other or shared a case or referral with another child (e.g., Child A and B were not on the same case, but both shared a case with Child C at some point). Because case IDs often follow a household or primary caregiver, this formulation may ignore siblings who did not previously reside together and may inaccurately characterize unrelated children living in the same home as siblings.

Our target sample is children (ages 0-17 at the point of removal) who spent time in foster care during the observation period and had at least one sibling under age 18 at the date of their first removal episode of the observation period (N=2,583). From that sample, we excluded a small number of children who had invalid/missing data on key characteristics, as well as youth who entered foster care through the juvenile justice system or were identified as a teen parent with a child in foster care. The final sample was 2,297 children (4,260 child-placements).

Measurement

Placement Instability.

The first outcome is any move to a new foster care placement, irrespective of the reason why. Second, following prior work (e.g., Font, Sattler, & Gershoff, 2018), we measure non-progress moves using caseworker-recorded reason for placement end (see Appendix A). We caution that caseworkers can only select one reason and may differ in how they attribute placement endings. Non-progress moves include placement changes for reasons indicative of a problem with the fit or experience of the prior placement, such as allegations of maltreatment against the child, the foster parent or child requesting that the child be moved, or the placement being unable to meet the child’s physical or behavioral health needs. We also count runaway/abscondence as a form of non-progress move (1.2% of placement ends), as running away may signal that the placement is ill-suited to the youth’s needs (Sattler et al., 2018). The measure of non-progress moves excludes moves attributed to the availability of a more preferred placement (i.e., progress moves)¹ and exits from care. Ambiguous placement end reasons (4% of placement ends) were counted as non-progress moves.

Identifying Sibling Groups and Sibling Separation.

The primary predictor of interest is sibling separation. Siblings may be separated when they are placed in different foster homes/facilities in foster care, or when only part of a sibling group is in foster care. Using household rosters across all of a child’s CWS reports and open cases during the observation period, we first identified the total number of siblings a child had who were under age 18. Then, for each week that any member of the sibling group was in foster care, we used foster care entry and exit date to determine the number of siblings in foster care. Lastly, for children in foster care on a given week, we determined whether they were in the same placement as some, none, or all of their siblings using the agency’s unique foster care provider (facility or home) identifier.

Our primary measure considered in/out of foster care separations and within-foster care placement separations, coded into four scenarios: (1) Child alone in foster care: child is in foster care but no known siblings are in foster care; (2) Child alone in placement: child and at least one sibling are in foster care, but no siblings are in the same placement as child; (3) Partially with siblings: child is in the same foster care placement as at least one but not all siblings (other siblings may be in other foster care placements or not in foster care); and (4) With all siblings: child and all known siblings are in the same foster care placement.

Two alternative measures of sibling separation focus on children and their siblings who are in foster care, ignoring separations that occur when a child’s sibling remains in their home of origin or exits before the child. These measures underestimate experiences of sibling separation but may be more reliable if there are errors in how we characterize sibling groups. Alternative Measure 1 differentiates complete and partial placement separation, following prior work (Hegar & Rosenthal, 2011): (1) Child is alone: placed with no siblings; (2) Partially with siblings: child has at least one sibling in their placement and at least one sibling in another placement; and (3) Child placed with all siblings. Alternative Measure 2 considers the siblings history of placement together or apart, following Leathers (2005): (1) Continuously placed alone: placed apart from all siblings, not previously placed together; (2) Inconsistently placed alone: child placed apart from all siblings, previously placed together; and (3) Child placed with at least one sibling.

Covariates.

Our models adjust for characteristics of children, sibling groups, removals and placements that may influence both sibling separation and placement instability. Child characteristics are age, sex, and race/ethnicity. Sibling group characteristics are number of identified siblings, age of oldest sibling, and an indicator of whether a child had only same sex siblings. Removal episode characteristics are removal year (range = 2015-2019) and removal reasons. Removal reasons are seven non-mutually-exclusive indicators condensed from a larger variable set: child-related reasons (e.g., child behavior problems or disability); neglect (supervision, medical, and physical needs); abandonment or relinquishment; abuse (physical, sexual, or emotional); parent-child conflict; parent alcohol or drug abuse; and other. Placement characteristics are number of previous placements, setting (non-relative, kinship, congregate), and placement level (not assigned, mild, moderate, serious/severe).²

Analytic Approach

We use survival models to estimate associations between sibling separation and the hazard of a placement move, adjusting for child, sibling, removal episode, and placement characteristics. Survival analysis is an appropriate choice for our research questions because, although we are not interested in the timing of placement moves per se, analytic approaches that ignore timing may be biased because a substantial minority of placements are ongoing at the end of the observation period (i.e., data are right-censored). Our models use the Weibull survival distribution, which produced the highest goodness-of-fit score when compared with other parametric functions such as exponential, log-normal, or Gompertz distribution.

Children in the sample may experience multiple changes in placement, which is typically addressed with a multiple failure model. However, most placement moves occur within a few months of the start of the placement (Wulczyn, Kogan, & Harden, 2003), with the hazard of a placement move decreasing over time since the start of a placement. Consequently, placement time, rather than child-time, is most important to the hazard function and we model our observations at the child-placement (rather than child) level. Thus, time=0 at the start of each unique placement and ends when a child exits, moves, or (if the placement is ongoing) reaches the end of the observation period. Our models allow for a single failure event only (i.e., each child-placement can end only once) but we include child’s number of prior placements as a covariate. Because sibling separation may vary within a child’s placement (one child may change placements or exit care without a change in their sibling’s placement), we organize our data by placement-week, such that sibling separation is time-variant. Finally, our data are nested: placements nested within children, and children within sibling groups. Thus, our main results are based on three-level random intercept models, with Level 1 (L1) = placement, L2 = child, and L3= sibling group. Appendix E describes and presents additional alternate specifications of our models, the results of which were substantively consistent with our main models.

Results

Sample Description

An overview of sample characteristics is found in Table 1. Demographics and removal episode characteristics are reported at the child level (2,297 children), and placement characteristics are reported at the child-placement level (4,260 placements).

Table 1.

Sample Description; Full sample, Percent or Mean (SD)

Child-Level Characteristics (N=2,297 children)		Placement-level Characteristics (N=4,260 child-placements)
Demographics		Placement type
Age at removal		Kinship Care	56%
Younger than 2	18%	Foster Care	34%
2 to 4	21%	Congregate Care/Other facility setting	10%
5 to 7	21%	Placement Level
8 to 10	16%	Mild	74%
11 to 13	13%	Moderate	16%
14 or older	11%	Serious/severe	3%
Sex		Not assigned	7%
Male	50%	Placement end category
Female	50%	Non-progress move	23%
Race		Progress move	18%
White	35%	No move; exit	39%
Black	55%	No move; placement ongoing	20%
Hispanic	8%	Median length of placement (Weeks)	16.57 (38.97)
Other/Missing	2%
Sibling Characteristics
Number of siblings
1	44%
2	30%
3	15%
4 or more	11%
Oldest sibling’s age	9.82 (4.55)
Single gender sibling group	35%
Mixed gender sibling group	65%
Removal Episode Characteristics
Removal reasons
Child-related	5%
Neglect	43%
Abandonment and relinquishment	5%
Abuse	13%
Parent-child conflict	7%
Parent AODA	38%
Other	28%
Year Removed
2015	18%
2016	20%
2017	24%
2018	19%
2019	19%
Total time in care at end of observation period
<12 months	55%
12-24 months	26%
>24 months	19%

Child, sibling group, and removal characteristics.

The sample was relatively young (60% ages 7 or younger at the time of removal) and evenly split by sex. The majority of children were Black (55%); 35% were white, 8% were Hispanic, and 2% were another race/ethnicity or of unknown race/ethnicity. A plurality of children had 1 sibling (44%), though 11% had 4 or more; the majority (65%) of children had at least one sibling of a different sex. The most common removal reasons were neglect (43%) and parent alcohol or drug abuse (38%), and the majority of children were observed in care for less than 12 months (55%).

Placement characteristics.

Most placements (56%) were in kinship care, with 34% in non-relative family foster care and 10% in congregate care. Most placements were assigned as mild levels of care (74%). Approximately 41% of placements ended in a move (23% non-progress move; 18% progress move), 39% ended with an exit from care, and the remaining 20% were ongoing at the end of the observation period. Average placement length was 16.57 weeks.

Representativeness of Sample.

Compared with the national foster care population, our sample is comprised of a larger proportion of Black children and smaller proportions of white and Hispanic children. A larger proportion of our sample is in kinship care (national rate ~32%; USDHHS, 2019). We also compare the Allegheny County cohort to the national metric for placement stability (the proportion of children experiencing no more than two placements) in Appendix B. For children in care less than 24 months, the rate of placement stability in our sample falls between the 25^th and 50^th percentiles of all states. For children in care longer than 24 months, stability in our sample is higher than at least 75% of states; however, this is likely because our data constitute an entry cohort rather than all children in care (thus we observe a smaller proportion of long-term episodes than are used to compute the national statistics).

Sibling Separation Description

Table 2 shows the percentage of child placement-weeks where siblings were separated for each of the three separation measures. Using our primary measure, we calculate that children were placed with all identified siblings in nearly half (47%) of placement-weeks; rates were higher for children with only one sibling (61%) versus children with multiple siblings (35%). When siblings were not all placed together, children were still placed with at least one sibling in about 20% of placement-weeks. In approximately 15% of placement-weeks, children were alone in foster care, and in 17% of placement-weeks, they were alone in placement.

Table 2.

Sibling Placement Measures (by week)

		Total		Children with 1 sibling		Children with multiple siblings
Label	Detail description	N child-week	%	N child-week	%	N child-week	%
Primary measure
Sample = All children in foster care with siblings
Alone in foster care	Child is in foster care but none of child’s siblings are in foster care	21,431	15.02	14,535	22.03	6,896	8.99
Alone in placement	At least one sibling in foster care; no siblings in child’s same placement	24,715	17.32	11,373	17.23	13,342	17.39
Partially with siblings	Child is placed with at least one sibling; at least one sibling is either not in foster care or not in same foster care placement	29,356	20.57	--	--	29,356	38.25
With all siblings	Child is in the same foster care placement as all of their known siblings	67,226	47.10	40,082	60.74	27,144	35.37
Alternative 1 (ignoring siblings not in foster care)
Sample = Children in foster care with at least one sibling who is also in foster care
Placed alone	Placed with no siblings	24,715	20.38	11,373	22.1	13,342	19.1
Partially with siblings	Child is placed with at least one sibling and at least one sibling is in a different foster care placement.	19,609	16.17	--	--	19,609	28.08
With all siblings	Of those in foster care, all members of the sibling group are placed together	76,973	63.46	40,082	77.9	36,891	52.82
Alternative 2 (ignoring siblings not in foster care)
Sample = Children in foster care with at least one sibling who is also in foster care
Consistently placed alone	Child is currently placed apart from all siblings and was not previously placed with siblings	30,272	21.21	19,196	29.09	11,076	14.43
Inconsistently placed alone	Child is currently placed apart from all siblings, but was previously placed with at least one sibling	15,874	11.12	6,712	10.17	9,162	11.94
Currently with siblings	Child is currently placed with some or all siblings in foster care	96,582	67.67	40,082	60.74	56,500	73.63

Using Alternative Measure 1, where siblings not in foster care are not included in the separation measures, a much larger proportion of placement-weeks (63%) are identified as with all siblings. Using Alternative Measure 2, we observe that placement alone (with no siblings) was typically consistent rather than temporary -- there were few placement-weeks spent alone after a period of placement together (11%). Children with multiple siblings spent fewer placement-weeks consistently alone (14%), and more weeks with at least one sibling (73%) than children with only one sibling (29% and 61%, respectively). A breakdown of sample characteristics by sibling separation measure is found in Appendix C.

Hazard Model Results

Table 3 depicts results of multilevel survival models of the hazards of any placement move and a non-progress placement move. Coefficients are presented as hazard ratios (HRs), where a value of 1 indicates equal hazard of the failure event (i.e., placement move) across levels of the predictor variable; values less than 1 indicate a negative association (decreased hazard) and values greater than 1 indicate a positive association (increased hazard).

Table 3.

Mixed-effect Weibull Proportional Hazard Model Results, Primary Measure

	Any move		Non-progress move only
	H.R.	95% C.I.	H.R.	95% C.I.
Sibling separation (reference = with all siblings)
Alone in foster care	1.234*	[1.005-1.516]	1.356**	[1.085-1.696]
Alone in placement	1.644***	[1.380-1.959]	1.386**	[1.133-1.695]
Partially with siblings	1.303**	[1.069-1.588]	1.025	[0.807-1.301]
Child Characteristics
Age (reference = younger than 2)
2-4	1.096	[0.858-1.401]	1.075	[0.800-1.445]
5-7	1.165	[0.906-1.499]	1.394*	[1.050-1.851]
8-10	1.342*	[1.014-1.775]	1.345†	[0.984-1.839]
11-13	1.327†	[0.969-1.819]	1.545*	[1.104-2.162]
14 or older	1.352	[0.938-1.950]	1.641*	[1.118-2.408]
Sex (reference = male)
Female	1.004	[0.871-1.156]	1.018	[0.876-1.183]
Race (reference = White)
Black	1.256**	[1.064-1.484]	1.249*	[1.044-1.494]
Hispanic	1.619***	[1.236-2.122]	1.565**	[1.188-2.062]
Other/Missing	0.558*	[0.326-0.955]	0.604†	[0.332-1.098]
Sibling Characteristics
Number of siblings (reference = 1)
2	0.903	[0.747-1.091]	0.960	[0.786-1.173]
3	0.872	[0.679-1.119]	1.003	[0.773-1.302]
4 or more	1.282†	[0.969-1.694]	1.045	[0.780-1.399]
Oldest sibling’s age	1.027*	[1.002-1.052]	1.047***	[1.019-1.075]
Mixed gender sibling group	0.984	[0.835-1.159]	1.010	[0.850-1.200]
Removal Episode Characteristics
Removal reasons
Child-related	0.755†	[0.546-1.043]	1.064	[0.786-1.441]
Neglect	1.230*	[1.041-1.453]	1.096	[0.919-1.308]
Abandonment and relinquishment	0.807	[0.580-1.124]	0.860	[0.612-1.210]
Abuse	1.411**	[1.121-1.775]	1.221	[0.960-1.553]
Parent-child conflict	1.187	[0.906-1.555]	1.398*	[1.077-1.814]
Parent AODA	0.901	[0.761-1.067]	0.897	[0.746-1.078]
Other	1.180†	[0.990-1.407]	0.928	[0.764-1.128]
Year Removed (reference = 2015)
2016	1.015	[0.811-1.269]	0.936	[0.740-1.184]
2017	0.941	[0.758-1.170]	0.930	[0.740-1.169]
2018	1.148	[0.916-1.439]	1.205	[0.952-1.525]
2019	1.148	[0.896-1.469]	1.143	[0.866-1.508]
Placement Characteristics
Placement type (reference = kinship care)
Foster Care	2.981***	[2.574-3.451]	1.701***	[1.443-2.003]
Congregate Care/Other facility setting	3.559***	[2.857-4.433]	1.218	[0.933-1.590]
Placement Level (reference = mild)
Moderate	1.021	[0.864-1.207]	1.294**	[1.080-1.550]
Serious/severe	0.755	[0.523-1.089]	1.058	[0.725-1.545]
Not assigned	1.492***	[1.187-1.876]	1.111	[0.826-1.494]
Number of prior placements (reference = none)
1	0.484***	[0.412-0.570]	0.887	[0.740-1.064]
2	0.304***	[0.240-0.384]	0.761*	[0.589-0.984]
3	0.320***	[0.238-0.430]	0.921	[0.656-1.295]
4 or more	0.243***	[0.181-0.326]	0.773	[0.541-1.103]
Shape parameter (p)	0.787***		0.740***
Number of observations	142728		142728
Number of placements	4260		4260
Number of failures	1700		957
Wald chi-squared (df)	432.863(37)		251.444(37)

Notes:

^†0.1,

^*0.05,

^**0.01,

^***0.001.

All three separation groups (alone in foster care, alone in placement, and partially with siblings) were associated with a significantly higher hazard of any move. Compared with children placed with all of their siblings, the increase in hazard of any move was 23% for children alone in foster care, 64% for children alone in placement, and 30% for children partially with siblings. However, only children separated from all of their siblings (either alone in foster care or alone in placement) faced heightened risk of a non-progress move. Children placed partially with siblings and children placed with all siblings had statistically equivalent hazards.

Children over age 4 faced an increased hazard of non-progress moves, but age differences in the hazard of any move were smaller and not consistently statistically significant. Black and Hispanic children were more likely to experience any move and a non-progress move. The hazard ratio for Black children implies a roughly 25% increase in the hazard of both any move and specifically non-progress moves, compared with white children; for Hispanic children, the increase in hazard exceeds 50%. Males and females had equal hazards on both outcomes.

The hazard of any move was roughly three times higher (HR=2.981) in non-relative foster care compared with kinship care, but the hazard of a non-progress move was about 70% higher (HR=1.701). Congregate care placements had a substantially higher hazard of ending with any move compared with kinship care, but no statistically significant difference in the hazard of ending in a non-progress move. The difference in placement setting coefficients suggests that progress moves account for a substantial portion of moves in non-relative foster care and congregate care. Age of child’s oldest sibling and child removal reason related to parent-child conflict were positively associated with the hazard of a non-progress move.

Subgroup Models by Number of Siblings.

Subgroup models (Table 4) indicate that sibling separation is associated with increased hazard of any move for both sibling pairs and larger sibling groups. However, the positive association between sibling separation and a non-progress move is specific to children with one sibling. Among children with multiple siblings, there is no association between any category of sibling separation and non-progress moves.

Table 4.

Subgroup Models, Primary Measure

	Any move		Non-progress move only
	H.R.	95% C.I.	H.R.	95% C.I.
Panel A. Children with 1 known sibling
Sibling separation (reference = with all siblings)
Alone in foster care	1.154	[0.899-1.481]	1.482**	[1.121-1.958]
Alone in placement	1.693***	[1.337-2.143]	1.610***	[1.222-2.121]
Wald chi-squared (df)	209.871(33)		157.071(33)
Panel B. Children with >1 known sibling
Sibling separation (reference = with all siblings)
Alone in foster care	1.357†	[0.957-1.923]	1.213	[0.829-1.776]
Alone in placement	1.623***	[1.254-2.102]	1.209	[0.895-1.633]
Partially with siblings	1.381**	[1.102-1.731]	0.977	[0.745-1.282]
Wald chi-squared (df)	259.990(36)		128.896(36)

Notes:

^†0.1,

^**0.01,

^***0.001. Models include all controls from Table 3 (coefficients not shown).

Alternative Measures of Sibling Placement.

Alternative models (Table 5) consider only siblings of children who are in foster care. Thus, children alone in foster care are excluded and a placement-week is categorized as with all siblings if all of a child’s siblings who are in foster care in that week are in the same placement as the child. The first set of models (Alternative Measure 1) are consistent with findings in Tables 3 and 4: findings are robust to the exclusion of siblings not in foster care at the same time as the focal child. The second set of models (Alternative Measure 2) focus on the consistency of joint or separate placement (again, only considering siblings who are also in foster care). Being placed consistently alone or inconsistently alone are both associated with increased hazard of any move and a non-progress move. Again, associations are more consistent (both in magnitude and statistical significance) for children with one sibling versus multiple siblings.

Table 5.

Alternative measures of sibling separation

	Any move		Non-progress move only
	H.R.	95% C.I.	H.R.	95% C.I.
Models Using Alternative Measure 1 (reference = placed with all siblings in foster care)
Panel A. All children with a sibling in foster care
Placed alone	1.555***	[1.302-1.857]	1.398**	[1.137-1.719]
Partially with siblings	1.287*	[1.041-1.590]	1.176	[0.908-1.525]
Wald chi-squared (df)	399.990(36)		166.204(36)
Panel B. Children with one sibling in foster care
Placed alone	1.732***	[1.328-2.257]	1.593**	[1.184-2.144]
Wald chi-squared (df)	172.190(32)		100.563(32)
Panel C. Children with multiple siblings in foster care
Placed alone	1.469**	[1.151-1.873]	1.292†	[0.964-1.731]
Partially with siblings	1.327*	[1.060-1.661]	1.175	[0.889-1.552]
Wald chi-squared (df)	249.562(35)		106.153(35)
Models Using Alternative Measure 2 (reference = placed with at least one sibling)
Panel A. All children with a sibling in foster care
Consistently placed alone	1.322**	[1.115-1.568]	1.335**	[1.106-1.612]
Inconsistently placed alone	1.432***	[1.175-1.745]	1.410**	[1.132-1.756]
Wald chi-squared (df)	420.515(36)		252.308(36)
Panel B. Children with one sibling in foster care
Consistently placed alone	1.296*	[1.028-1.633]	1.432**	[1.103-1.861]
Inconsistently placed alone	1.729***	[1.275-2.343]	1.824***	[1.305-2.551]
Wald chi-squared (df)	203.377(33)		159.339(33)
Panel C. Children with multiple siblings in foster care
Consistently placed alone	1.315*	[1.027-1.684]	1.280†	[0.970-1.688]
Inconsistently placed alone	1.218	[0.934-1.589]	1.165	[0.860-1.576]
Wald chi-squared (df)	250.937(35)		128.363(35)

Notes:

^†0.1,

^*0.05,

^**0.01,

^***0.001. Models include all controls from Table 3 (coefficients not shown).

Subgroup Analyses by Race/Ethnicity.

We also examined whether our findings were consistent across racial groups. In most jurisdictions, Black children are overrepresented in foster care, reflecting higher rates of maltreatment (Sedlak, McPherson, & Das, 2010) and associated risk factors (Drake, Jonson-Reid, Kim, Chiang, & Davalishvili, 2020), and, in some instances, differential treatment (Pryce et al., 2019). Some also claim that foster care is disproportionately harmful to Black children, though there is limited research to address this question (Barth et al., 2020). Our descriptive analysis (Appendix C) shows that Black children are more likely to be placed partially with siblings, but less likely to be placed with all siblings, than white or Hispanic children. Because Black children on average spend longer lengths of time in foster care and are more likely to be placed with families of a different racial or cultural background than themselves, separation from siblings may be particularly negative. Our subgroup models by race (found in Appendix D) provide modest support for the hypothesis that associations between sibling separation and placement moves are stronger for Black children. Specifically, whereas there is no statistically significant association between being alone in foster care and the hazard of any move for white children (HR=1.092, p>.1), associations are large, positive, and statistically significant for Black children (HR=1.535, p<.001). This difference is also found for non-progress moves specifically. Findings are less conclusive for being alone in placement. For any move, the pattern holds (Black children, HR=1.745, p<.001; white children, HR=1.374, p<.1). Yet, for non-progress moves, the alone in placement coefficient is nearly identical for Black and white children (HR=1.300 versus 1.330, respectively) and neither coefficient is statistically significant at the p<.05 level. Associations between placement partially with siblings and the instability outcomes are similar for Black and white children. The sample of Hispanic children is rather small. However, we find a null or marginally negative association between being alone in foster care and placement moves for Hispanic children, and strong positive associations of being alone in placement with any move (HR=2.297, p<.01) and with non-progress moves (HR=2.502, p<.01).

Discussion

Despite longstanding federal and state policies promoting sibling placement for children in foster care, rigorous research on the outcomes of sibling separation remains limited. We investigated associations between sibling separation and placement instability, using both a generic measure of instability (any move) and a narrower measure intended to capture a placement disruption that did not facilitate a specific goal or preference for the child (non-progress move). Because preventing non-progress moves is a greater system priority – and likely more relevant to improving children’s experiences in foster care—we place primary emphasis on those findings. We find that sibling separation is positively associated with both measures of placement instability. This is largely consistent with the work of prior scholars, who have argued sibling separation may compound other relational losses in foster care (McCormick, 2010). Yet, partial separation (placement with some but not all siblings) is not associated with a non-progress move. This may suggest that, when all members of a sibling group cannot be placed together, the benefits of placing each child with at least one sibling may provide similar benefits. Although policy requires efforts to keep all siblings together, this is often not feasible for large sibling groups and evaluation metrics should measure and acknowledge partially-together placements.

Furthermore, separation from siblings was positively associated with a non-progress move for children with one sibling but not for children with multiple siblings. Although we do not know of other studies that have compared children with one versus multiple siblings, we pose two possible explanations for why sibling separation could diminish placement instability more so for children with one sibling. First, larger size of sibling group appears to adversely impact children’s perceptions of the quality of parenting received (Kidwell, 1981) and children with more siblings tend to fare worse on a variety of intellectual measures (Downey, 2001). This is most commonly explained by dilution theory (Blake, 1986), which argues that parents have a finite amount of time and emotional energy for their children, and thus this investment is diluted when there are more children, even where there is no scarcity of economic resources. In the context of foster care, the dilution of resources would be affected by the number of siblings within the same placement, rather than the total number of siblings a child had. Second, children with only one sibling may experience less sibling conflict or competition between siblings than children in larger sibling groups (Newman, 1996), which may imply fewer challenges in providing safe and stable foster care for a sibling pairs versus a larger sibling group. Yet, studies also suggest that larger sibling groups are more emotional connected and may develop more favorable personality traits (Newman, 1996). This would seem to suggest specific harms from separating larger sibling groups, but may also suggest that children in larger sibling groups are more adaptive to different environments and have more developed social skills that enable them to connect with other (non-related) children or adults in their foster home.

In models where we consider consistency of separation, we find that these associations persist even when separation is long term. That is, children consistently placed alone versus children who were previously placed with siblings and then separated have statistically equivalent hazards of moves. Thus, it is incumbent on agencies to continue to pursue opportunities for siblings to be placed together even if they are unable to be placed together at the outset of their placement episode. At the same time, moves for the purpose of sibling placement may be harmful when it requires a child to be separated from a long-term caregiver. Targeted recruitment of families that are open to providing care for sibling groups is likely the best means of avoiding sibling separation immediately upon entry. Although there is limited research on the characteristics of foster families able or willing to provide for sibling groups, adequate care for multiple children may be more attainable in the context of two-parent households and households where at least one parent has a flexible or reduced-hours work schedule to avoid dilution of resources. Another option would be for agencies to recruit widely, but where possible to “reserve” foster homes open to multiple children for sibling group care, rather than placing multiple unrelated children in those homes.

We also find stronger associations between separation and instability for Black children compared with white children. This may reflect differential selection into separation; for example, if white siblings are separated mostly due to safety issues and Black siblings are separated due to logistical or practical constraints (e.g., fewer placement options), we would expect white children to experience less harm from separation. Alternatively, there may be unobserved differences in the closeness or quality of sibling relationships. More research on the context of separation and sibling relationship quality is needed to verify and explain these patterns. Regardless, as CWS agencies increase their focus on issues of racial equity, they may consider options to improve rates of sibling placement and mitigate impacts of separation.

Although our findings suggest that sibling placement is beneficial for children, we caution that agencies already, due to both legal and best practice guidelines, attempt to keep siblings together. Without additional resources or supports, further progress in keeping siblings together may be difficult. Many agencies struggle with foster home recruitment and retention (Government Accountability Office, 2015) or are not able to effectively use their available foster homes (Wulczyn et al., 2018). This may leave an inadequate supply of foster parents with the commitment and skills to accept and succeed with sibling placements. To this end, the emergence of evidence-based practices for improving sibling relationships among children in foster care (Kothari et al., 2017) is a critical development. Providing services that support and improve sibling relationships may increase the potential for co-placement to be safe, stable, and beneficial to all siblings, and ease the strain on foster families providing care for sibling groups. Moreover, improved use of technology to match children and foster families (Font, 2020) may facilitate additional progress in both the rates and outcomes of sibling placement.

We note several limitations to this study. The data come from a single urban county and findings may not generalize to other areas. Our data include limited information on children’s special needs (e.g., physical, emotional, and behavioral challenges), which may inform decisions to separate siblings, constrain the number of placement options for keeping siblings together, and influence foster parents’ levels of stress and ultimately their decisions to request that a placement be terminated. We cannot externally validate whether the child placement level, ranging from mild to intensive, accurately reflects a child’s needs or challenges. Improved measurement and documentation of children’s disability status and socioemotional functioning will greatly improve the quality of evidence on both sibling separation effects and placement instability broadly. Clear documentation of reasons for sibling separation remains essential to identify siblings separated for health or safety reasons versus logistical or resource barriers.

Another limitation is reliance on caseworker-report to ascertain reasons for placement ending, and thus develop our metric of non-progress moves. There is likely some degree of variability in caseworker coding schemes. However, because we only sought to separate out progress moves (not to delineate causes of placement disruption in finer detail), we believe this is a minor limitation. Progress moves are relevant to understanding whether and how agencies are meeting federal performance objectives (e.g., the Child and Family Services Reviews) and thus are likely to be coded as such where applicable. In addition, unlike studies conducted elsewhere (Font et al., 2018), our data have few cases coded with ambiguous reasons (e.g., “Other”).

Our data also contain limited information on placement characteristics that may affect placement disruptions, independent of sibling separation (e.g., inadequate training; financial hardship; changes in the foster family context, such as pregnancy, illness, or divorce). Improved collection and integration of data on foster and kinship environments is greatly needed to better understand the role of sibling separation in placement disruption – particularly family and relationship dynamics that may mediate or moderate sibling placement effects. Like other observational studies, our findings should not be interpreted causally. Ongoing federal efforts to improve the quality and reliability of CWS data (Administration for Children and Families, 2016) may allow for more rigorous national studies in the future.

Lastly, our data contain no information on sibling visitation. Visitation may mitigate some negative impacts of sibling separation, particularly for children who are in foster care for short periods of time and exit to reunification. Federal policy requires visitation when siblings are in foster care but not in the same placement, except in limited circumstances where it would be harmful. When some siblings are in foster care and others are not, contact is encouraged but cannot be legally mandated. Consequently, depending on the nature of sibling separation, the degree of contact likely varies (and thus, effects of sibling separation may vary too).

Notwithstanding these limitations, our study finds consistent evidence that sibling separation is associated with increased risk of placement moves for children in foster care, including non-progress moves (which may indicate problems with children’s functioning or connection to caregivers). Our findings suggest that separation is especially negative for children who have only one known sibling, and that for larger sibling groups, placement with at least one sibling produces similar benefits as placement with all siblings. In total, our findings provide modest support for current policies that promote sibling placements.