A Randomized Controlled Trial of Family Finding

Abstract

Background:

Youth who have experienced foster care are at risk of negative outcomes in adulthood. The family finding model aims to promote more positive outcomes by finding and engaging relatives of children in foster care in order to provide options for legal and emotional permanency.

Objectives:

The present study tested whether family finding, as implemented in North Carolina from 2008 through 2011, improved child welfare outcomes for youth at risk of emancipating foster care without permanency.

Research Design:

A randomized controlled trial evaluation was carried out in nine counties in North Carolina. All children eligible for intervention services between 2008 and 2011 underwent random assignment. Effects were tested with an intent-to-treat design. Outcome data were obtained for all subjects from child welfare administrative data. Additional outcome data for a subset of older youth came from in-person interviews.

Subjects:

Subjects included 568 children who were in foster care, were 10–17 years old (at time of referral), had no identified permanent placement resource, and had no plan for reunification.

Measures:

The confirmatory outcome was moves to more family-like placements, whether through a step-down in foster care placement or discharge from foster care to legal permanency.

Results:

No impact on the confirmatory outcome was observed. Findings regarding exploratory impacts are also described; these must be interpreted with caution, given the large number of outcomes compared.

Conclusions:

The evaluation failed to find evidence that family finding improves the outcomes of older youth at risk of emancipation from foster care.

Keywords

child welfare content area outcome evaluation (other than economic evaluation)design and evaluation of programs and policies foster care family finding

Social networks within families and communities are important for the development of positive social skills and self-identities (Hines, 1997). Such networks may be particularly critical during the transition to adulthood. Important cognitive, behavioral, and emotional development occurs between the ages of about 18 and 25 (Arnett, 2007; Arnett & Tanner, 2006, as cited in Avery, 2010). During this time, many young people remain at or return to their parents’ homes or rely on parents and other supportive adults for material and emotional assistance (Furstenberg, 2010; Swartz, Kim, Uno, Mortimer, & O’Brien, 2011).

For youth in foster care, time spent living apart from their families likely weakens their social networks (Perry, 2006); this may be a particular challenge as these youth transition to adulthood. The majority of youth are discharged from foster care to their birth families or other family members (U.S. Department of Health and Human Services [DHHS], 2014); many others return to their birth families upon aging out of foster care (Courtney & Dworsky, 2006). However, youth’s families may have limited capacity to support them, and many may have lost contact with extended family members who might otherwise have stepped in to help. Furthermore, youth have lost the formal supports of the foster care system during these years.

Understanding factors that can support the well-being of youth who have experienced foster care is important, due to their elevated risk for negative outcomes. Adults with a history of foster care tend to fare less well across multiple measures of socioemotional well-being, compared with those who had never experienced foster care, according to a nationally representative survey (Cook-Fuong, 2000). Especially vulnerable are youth who emancipate or age out of foster care, that is, exit without having achieved legal permanency through reunification with their family of origin or through adoption or guardianship with another family. These youth are at increased risk of negative outcomes in terms of homeless and/or housing instability (Courtney et al., 2011; George et al., 2002; Pecora et al., 2005), employment and income (Courtney et al., 2011; Dworsky & Courtney, 2000; George et al., 2002; Pecora et al., 2005), physical health (Courtney et al., 2011), and social and emotional functioning (Courtney et al., 2011; Pecora et al., 2005).

More than 23,000 youth emancipated to independence in 2013 (U.S. DHHS, 2014). Long spells in foster care are more strongly associated with a reduced likelihood of reunification than is a child’s current age or age of entry into care; however, entering care at older ages is negatively associated with the likelihood of adoption, according to a review of the literature (Leathers, Falconnier, & Spielfogel, 2010). An analysis of administrative child welfare data suggests that youth who linger in care are at elevated risk of emancipation (Maza, 2009).

The establishment of legal permanency for older youth and youth who linger in foster care might protect them against negative well-being outcomes. In addition, ensuring that they have enduring relationships with relatives and other supportive adults in place, regardless of legal arrangements—that is, “emotional permanency”—is another promising pathway to improve outcomes. Indeed, research indicates that strong social networks, foster care placements in family settings rather than congregate care (Perry, 2006), and stable family environments (Pecora et al., 2005) are all positively associated with well-being of youth who have experienced foster care. In addition, positive, regular contact with birth family members may contribute to improved social well-being and adjustment for these youth (Andersson, 2005).

Federal recognition of the importance of the connections between children in foster care and their kin was codified in the Fostering Connections to Success and Increasing Adoptions Act of 2008 (P.L. 110-351). Among other provisions, congress authorized US$75 million over 5 years for grants to enable states, tribes, and nonprofit organizations to implement programs with the explicit purpose of increasing permanency for children and youth, including relative search and engagement programs such as “intensive family finding.”

The goal of the family finding model, originally designed to serve youth lingering in foster care, is to find and engage relatives and other kin to provide options for legal and emotional permanency. The family finding model is comprised of six stages as developed by Kevin Campbell (2005, 2010a, 2010b): (1) discovering at least 40 family members and important people in the child’s life, (2) engaging as many family members and supportive adults as possible, (3) planning for the successful future of the child, (4) making decisions that support the legal and emotional permanency of the child, (5) evaluating the permanency plans developed for the child, and (6) providing follow-up supports to/for the child and his or her family.

Prior Evaluations of Relative Search and Engagement Programs

Early nonexperimental evaluations of family finding yielded promising findings for older youth who had lingered in foster care. For example, many youth served by the California Permanency for Youth Project (CPYP) experienced progress in the areas of legal permanency and permanent connections (CPYP, 2008, 2010; Wakcher, 2010). Permanent connections were similar to parent–child relationships but not necessarily legally formalized.

More recently, evaluators have completed 11 studies of family finding in additional sites funded through the 2009 Family Connection Discretionary Grants from the Children’s Bureau, an office of the Administration for Children and Families of the U.S. DHHS. Building on a cross-site evaluation report that summarized evaluation findings (James Bell Associates, 2013), Vandivere and Malm (2015) carried out an independent review of the grantees’ final reports. Vandivere and Malm focused on the subset of six that were experimental evaluations and added results from two privately funded experimental evaluations of family finding programs. Sample sizes in the studies ranged from 175 to 883. Three interventions (in addition to the intervention that is the focus of the present article) served youth who had lingered in care; two of these served children of any age and one served ages 4 to 16. None of the three studies found statistically significant impacts on legal permanency, and none examined relational permanency. A study in Maryland found that the intervention increased the likelihood of having independent living as a case goal; one in Wisconsin found that the intervention increased placement instability, decreased having placement with a relative as a case goal, and increased having guardianship as a case goal.¹

Among the remaining studies, which evaluated programs targeting either youth new to care or combined populations of youth new to care and already in care, only one showed favorable impacts on legal permanency (Vandivere & Malm, 2015). That study (Landsman, Boel-Studt, & Malone, 2014), carried out in Iowa and implemented by a private nonprofit child welfare agency, served children under age 18 referred for an initial foster care placement or a change in placement. With a relatively small sample size of 243, in comparison with the other studies, the Iowa study found that the intervention increased the likelihood of adoptions by relatives and reduced the likelihood of youth aging out of foster care. That same study was the only one to examine emotional permanency, and a positive impact was found.

An important caveat regarding the extant experimental evaluations pertains to cross-site variation in the intervention and evaluation designs. Furthermore, not all of the final reports are publicly available. Only two have been published in peer-reviewed journals to date: the Iowa study (Landsman et al., 2014), which had the most positive findings, and the Wisconsin study (Garwood & Williams, 2015). Among the available reports, not all provide sufficient information to judge the rigor of the analyses, and—due to our observation that some findings that are included in James Bell’s cross-site study do not appear in the individual grantee reports—we suspect that not all findings (and in particular, null findings) are included. These limitations hamper external assessment of the findings and leave large gaps in the knowledge base about the effectiveness of family finding as well as the contexts in which it is effective and features of model design that are more or less effective.

Purpose

Evidence is limited on the effectiveness of child welfare interventions for children who are at risk of aging out of the foster care system (Montgomery, Donkoh, & Underhill, 2006). For family finding in particular, the evaluation by Landsman, Boel-Studt, and Malone (2014) yielded the strongest evidence, but it remains unclear whether family finding is effective with older youth at risk of emancipation, in contrast with the Iowa study’s more heterogeneous population.

To add to the knowledge base, we carried out an evaluation of family finding in North Carolina. Given the short time in which to affect outcomes before youth reached the age of majority, as well as a limited follow-up time for the study, the key outcome for the present study is a move either to a less restrictive placement setting or from a nonrelative to a relative placement. Such moves represent a proximal outcome that may facilitate ultimately achieving emotional or legal permanency (although moves to adoptive or guardianship placements, if they did occur, were incorporated in the outcome measure). We theorized that the increased connections to kin identified and engaged through the intervention might enhance the well-being of children and youth, enabling them to make positive placement changes. For example, connection to a supportive adult might result in improved mental health and avoidance of risky behaviors, which might enable a youth in congregate care to live in a family setting, or might make a relative feel comfortable providing a home for a youth. In addition, stepping down to a family foster home, and in particular to a relative’s home, may provide an opportunity for the youth to develop permanent emotional connections with that family.

We selected a single outcome to be the focus of a “confirmatory” analysis because examining impacts on multiple outcomes results in an elevated likelihood of finding one or more significant impacts by chance. We examined additional outcomes, including those available only for a subset of the sample and for which statistical power to detect impacts was lower, as “exploratory,” an approach recommended by Schochet (2008).
The confirmatory research question addressed in this study is as follows: Compared with similar children not receiving family finding services, are those assigned to receive family finding services more likely to make positive moves to more family-like placements, whether through a step-down in foster care placement or discharge from foster care to legal permanency?

The exploratory research questions are as follows: Compared with similar children not receiving family finding services, are those assigned to receive family finding services more likely to experience positive child welfare permanency outcomes? Are they more likely to experience positive safety outcomes in terms of reallegations of abuse/neglect? Are they more likely to experience positive well-being outcomes?

Additional exploratory research questions include: Do program impacts across outcomes differ for children referred at ages 12 and under versus 13 and older? Do they differ for those referred to public agencies compared with those referred to private agencies? Impacts might differ by child age because of the short time until older youth reach the age of majority and, particularly for those who have lingered in care without achieving permanency, outcomes may be less malleable than for younger youth. In addition, impacts might vary by the type of agency implementing the intervention. Little is known about the consequences for child and family outcomes of child welfare privatization; the modes of collaboration between the public and private agency may play a role in effectiveness (Collins-Camargo, McBeath, & Ensign, 2011). Some evidence suggests that private, for-profit agencies may be more amenable to implementing innovative and/or evidence-based interventions than are public agencies (Aarons, Sommerfeld, & Walrath-Greene, 2009).

Method

We carried out an experimental outcome evaluation in which treatment group members were referred for family finding services in addition to traditional child welfare services, whereas the control group received traditional child welfare services only. A private foundation paid for both the intervention and corresponding evaluation. Participation in the evaluation by program providers was a prerequisite for receiving the program funds.

The foundation aimed to implement family finding in up to 10 counties that demonstrated a commitment to, and possessed the capacity for, implementation. Ultimately, nine counties, all within North Carolina, were included. A private agency operated the intervention in three counties beginning in July 2008, and the public child welfare agency launched the program in the remaining six counties in January 2009. In each county, a family finding specialist was responsible for implementing the intervention in conjunction with the child’s child welfare team. Specialists served approximately five to seven cases at a time, with average service periods of 5 months each. The intervention does not require direct youth participation (although in most cases, youth did actively participate; Malm, Vandivere, Allen, Williams, & McKlindon, 2014), so all treatment group youth were assigned to the caseload of a family finding specialist.

Sample and Data

The sample included children who were eligible and referred for intervention services. Children were eligible if they were in foster care, were 10 to 17 years old (at time of referral), had no identified permanent placement resource, and had no plan for reunification, or if they were the younger sibling of such a child and also lacked an identified permanent placement resource or plan for reunification.

Random assignment occurred from June 2008 through May 2011. Each family finding specialist waited until at least two eligible cases (either a child or sibling group) had been referred. The specialist then entered the names and client identification numbers for pairs of cases into an automated random assignment module that was part of the family finding case management database. The module then randomly assigned one case to the treatment group and the other to the control group.

Administrative data

We obtained outcome and covariate data from North Carolina’s child welfare information system in October 2012. As is frequently true with administrative data, we encountered some data quality problems, notably regarding children’s placement moves.² We also obtained state administrative data from the School of Social Work at the University of North Carolina. These data included many of the same variables used in our own analyses, including variables that had been cleaned by the University of North Carolina.³ As a test for the robustness of our findings to alternative assumptions on how to treat inconsistencies in the data, we replicated our analyses with the University of North Carolina’s data set.

Interview data

We obtained additional data through in-person interviews in English with the subset of study participants who were age 13 or older at study enrollment. We adapted the interview protocol from that used in the Multi-site Evaluation of Foster Youth Programs (Courtney et al., 2011). Youth also completed a pencil-and-paper module designed to assess behavior problems, the youth self-report (Achenbach, 1999). The hour-long interviews occurred where the child was living, which may or may not have been a foster care placement. We carried out two rounds of interviews from summer 2009 through May 2013 with response rates of 82.7% at 12 months following study enrollment and 77.8% at 24 months. Lack of response was generally due to inability to find youth.

Analytic sample

A total of 573 children underwent random assignment, and we obtained administrative child welfare data for all but five children. (See Table 1.) Of the 387 eligible for the survey, 305 completed the first interview, and 281 completed the second.⁴

Table 1.
Study Sample.

Stages of study enrollment and data collection Total Treatment Control

Underwent random assignment 573 295 278

Obtained child welfare administrative data 568 291 277

Eligible for interview (≥age 13 at study enrollment) 387 187 200

Obtained Round 1 interview data 305 143 162

Obtained Round 2 interview data 281 132 149

Measures

Data pertaining to youth well-being (other than safety) came from the youth interviews. All other outcomes, including the confirmatory outcome, came from the administrative child welfare data.

For the confirmatory outcome, we categorized placements settings in the following order from least to most restrictive: (1) discharge to legal permanency, including a finalized adoption, guardianship, or reunification; (2) parents’ home; (3) trial home visit; (4) relative’s home including relative adoptive home, living in home of relative, and relative family foster home; (5) specialized relative family foster home; (6) nonrelative’s home including nonrelative adoptive home, adoptive foster home, home of legal guardian, or family foster care home; (7) specialized nonrelative home including specialized family foster care home, therapeutic home, or emergency shelter; (8) small congregate care setting including residential school, maternity home, small residential group home, and small treatment group home; (9) independent living arrangement; and (10) large congregate care setting including large group residential or treatment facility, hospital, Department of Juvenile Justice and Delinquency Prevention, and jail, lockup, detention facility. Any move from a “higher” category to a “lower” category—when compared with the youth’s placement setting at the time of study enrollment—counted as a positive move (i.e., a positive outcome). Moves to placement types that we assumed to be temporary or that could not be categorized (specifically, children’s camp, runaway, respite, and other) were disregarded, because we could not consistently determine whether youth returned to their prior placement or not.

Exploratory outcomes are listed (along with analytical findings) in Tables 3 and 4. All outcomes were measured dichotomously, with the exception of the placement stability measure, which was a count, and the social support and self-efficacy measures, which were continuous.

Because the confirmatory outcome was a composite outcome that used information both on foster care placement changes and reasons for discharge from foster care, we included additional exploratory outcomes that disaggregated the confirmatory outcome. These include (1) discharge to permanency and (2) a positive foster care placement change. In addition, because the confirmatory outcome did not assess whether children maintained a positive move nor whether any negative moves occurred, additional exploratory outcomes assessed whether (1) the last move during the study period was to a more positive placement than the baseline placement setting, (2) any negative move compared with the baseline placement occurred, and (3) a negative move when comparing the last move with the baseline placement occurred. We also examined relative placements separately from overall placement changes with the following: (1) discharge from foster care to a relative, (2) last placement setting during the study period was with a relative, and—combining the discharge and placement data—(3) discharge from foster care to a relative or the last foster care placement setting was with a relative.

Social support was assessed based on responses to seven questions about how many different people the youth could rely on for material and emotional support, such as socializing, getting advice, borrowing money, or listening to the youth’s problems. Since responses for some questions tended to be higher than for others, numbers were standardized prior to summing into an index. Self-efficacy was assessed via an index based on responses to 18 questions about how prepared youth felt to handle various aspects of independent living such as living on their own, getting a job, obtaining housing, managing money, preparing meals, and dealing with legal and other problems. Youth responded to each item via a 5-point Likert-type scale. Responses were summed with total possible scores ranging from 18 to 72, with low scores indicating higher self-efficacy. If 5 or more items were missing, a summary score was not calculated. If between 1 and 4 items were missing, the summary score was weighted.

It should be noted that the substantive significance of impacts on child maltreatment reports can be difficult to interpret. Maltreatment reports might increase in conjunction with the implementation of an intervention due to an increased surveillance effect. That is, if families have more contact with program providers, then program providers have more opportunities to detect and report maltreatment, regardless of whether the actual incidence of maltreatment has changed. We hypothesized that family finding might reduce actual maltreatment by strengthening the family network. Thus, despite its limitations, we have included maltreatment reports as an exploratory outcome because no other data on maltreatment were available.

Sample Description

Demographic and case history characteristics, based on administrative data, are shown in Table 2. Overall, 38% of the youth were ages 16 or older, 23% had been in care for 5 or more years, and a third were in a congregate care setting at study enrollment. Taken together, the case history and demographic characteristics did not explain a significant amount of variation when regressed on the random assignment indicator, χ²(26) = 17.67, p = .8873.⁵

Table 2.
Characteristics of Children, by Experimental Group Membership.

Child characteristics Treatment (%) Control (%)

Child disabilities (n = 567)

Emotionally disturbed 18 15

Other diagnosed condition 15 18

Child is male (n = 567) 58 57

Child age at study enrollment (n = 567)

Age 0–12 26 18*

Age 13–15 38 42

Age 16 or older 36 40

Child race and Hispanic origin (n = 561)

Non-Hispanic White 38 43

Non-Hispanic Black 53 47

Hispanic 5 7

Non-Hispanic other 4 3

More than one entry into foster care (including the initial removal during the study period; n = 567) 19 23

Time in foster care prior to study enrollment (n = 564)

Less than 1 year 18 24

At least 1 year but less than 2 years 27 18*

At least 2 years but less than 5 years 33 35

5 or more years 22 23

Type of foster care placement setting at referral (n =553)

In foster care, living with relatives 4 1*

Nonrelatives family foster home 62 68

Congregate care setting 34 31

Removal reason (for episode of care during study period; n = 567)

Relinquishment/abandonment 7 12

Physical abuse 10 10

Sexual abuse 7 6

Drug or alcohol abuse: child 1 2

Drug or alcohol abuse: parent 19 18

Child behavior problems 15 22

Child disability 1 3

Parent inability to cope 25 24

Parental death 4 3

Inadequate housing 8 14

Parents incarcerated 2 4

Neglect 77 70

Among youth age ≥13 at study enrollment^a (unweighted)

Status of parents (n = 304)

Parents deceased and/or unknown if either is living 14 6

One parent is known to be living 27 31

Both parents are known to be living 58 63

School/health professional ever told youth he/she had learning disability (n = 304) 31 22

Youth ever previously ran away (n = 304) 45 39

^aData were collected 1 year postbaseline, so these incidents may have occurred after study enrollment.

p < .10. p < .05. p < .01.

Overall, the interview youth were generally similar to those who were not interviewed in terms of demographic characteristics, with the obvious exception of age. In addition, those in the interview sample were less likely to have been male (56% compared with 65%, p < .10). A number of case history characteristics did vary by age. At study enrollment, older youth were more likely than younger youth to be in a congregate-care placement (38% compared with 14%, p < .01), less likely to live with a nonrelative foster family (60% compared with 80%, p < .01), and slightly less likely to live with relatives (2% compared with 6%, p < .10). A larger share of the older youth had been in foster care for more than 5 years (25% compared with 15%, p < .05). Reasons for entry into care sometimes varied by age as well. Referral for child behavior problems was more common among older than younger youth (21% compared with 10%, p < .05), but other referral reasons were less common among the older youth (sexual abuse: 5% compared with 12%, p < .05; incarcerated parents: 2% compared with 7%, p < .05; and neglect: 7% compared with 9%, p < .05).

Within the interview sample, the treatment and control groups were generally similar, with the exception that a larger share of control group youth than the treatment group youth had entered care due to relinquishment or abandonment (13% compared with 7%, p < .10). Also, the shares of youth living in relative or nonrelative family homes at the time of study enrollment varied by experimental group but only by two percentage points or less.

A few sample statistics from the interview data are also shown in Table 2. In order to provide a description of sample members who responded to the survey, including any differences between treatment and control group sample members, the table presents unweighted statistics. Among all the older youth, 26% had learning disabilities, and only 61% knew that both parents were alive. Some of these circumstances reported in the interviews may have arisen in the year following study enrollment. Nevertheless, the relatively high percentages of youth whose parents were not known to be living, who had spent time in congregate care settings, who had multiple spells in foster care, and who had run away while in care, as well as the long-time spent in foster care, indicate that the target population was vulnerable and had experienced challenging circumstances above and beyond their histories of maltreatment.

Analysis Methods

To estimate program impacts, we carried out intent-to-treat analyses (ITT) in which we compared the outcomes for children assigned to the intervention with those for children not assigned to the intervention. Such analyses maintain the statistical similarities of the treatment and control group, supporting our ability to attribute causality for any observed impacts on outcomes to assignment to the intervention. Fortuitously, we found little evidence of random assignment contamination; cross-overs (i.e., control group members receiving intervention services) occurred only for three children.

Experimental analyses should include covariates that may be predictive of outcomes of interest, even when statistically significant differences in potential covariates do not emerge across the experimental groups (Egbewale, 2015; Knol, Groenwold, & Grobee, 2012). Accordingly, we controlled for time in care at study enrollment as well as indicators for behavioral/emotional disability, any other disability, male gender, more than one spell of foster care, non-Hispanic White race, removal due to neglect, removal due to inadequate housing, placement in congregate care at baseline, and placement in nonrelative foster home at baseline. Covariates for the well-being models additionally included whether or not one or both of the youth’s parents were deceased. We also tested the sensitivity of our results with models that included controls for county.

Due to sporadic missing data on covariates, we carried out a single-imputation approach—dummy variable adjustment. Specifically, we imputed missing covariates with mean values and included dummy variables to indicate missingness on each covariate. While dummy variable adjustment typically introduces bias into coefficient estimates (Allison, 2009), the bias in experimental impact estimates is comparable to bias resulting from the use of more sophisticated methods (Puma, Olsen, Bell, & Price, 2009). As Puma and colleagues point out, the experimental assignment indicator is never missing, and the process of randomization should result in no correlation between it and other independent variables. To test for sensitivity of the results to this approach, we also imputed missing data using Stata’s ice (imputation by chained equations) procedure (Royston, 2005), generating five plausible values for each missing value.

In contrast to the administrative data, nonresponse was a problem with the interviews. To reduce nonresponse bias, we developed a set of weights as recommended by Deke and Puma (2013). To do this, we regressed nonresponse on each youth’s experimental assignment group as well as on variables from the administrative data including presence of a diagnosed mental health condition or of any other diagnosed condition; gender; age in years; total number of foster care spells; whether the youth entered foster care due to neglect or housing issues, non-Hispanic White race; whether the youth was in a congregate care setting, with a nonrelative, or with some other setting at random assignment; the length of time the youth had been in foster care; and county. These covariates explain 9.6% and 8.6% of the variation in nonresponse in the 12- and 24-month surveys, respectively, according to likelihood-based pseudo R ² provided by SAS version 9.3. We then calculated the weight as the inverse of the predicted probability that each individual would be a nonresponder. We set the weight equal to 1 for the four youth who responded to the survey but who lacked any administrative data. Weights for both waves range from 1 to 2.89.

We estimated impacts with logistic regression models for all binary outcomes, linear models for continuous outcomes, a Poisson models for the number of placements, and multinomial logistic regression for the type of living arrangement 12 and 24 months postbaseline. Models using the administrative data were multilevel in order to account for the nesting of children in counties and in sibling groups. The field has not come to consensus regarding the appropriate approach for weighting multilevel models with categorical data (Gelman, 2007). Therefore, we accounted for clustering at only one level (sibling group) in the weighted interview data by estimating fixed-effects models with robust standard errors using Stata’s vce option, which specifies how to estimate the variance-covariance matrix (VCE) corresponding to the parameter estimates. We used the xtmelogit, xtmepoisson, and xtmixed commands in Stata 13.1, respectively, with the unweighted administrative data, and the logit, regress, and mlogit commands with the weighted interview data. For the administrative data, we tested the sensitivity of the results to the method of accounting for nesting of the data, by adjusting for clustering at the sibling group–level only.

Because a single, confirmatory outcome is the focus of this study, no correction is necessary to the statistical significance of the impact on positive moves that would otherwise be required in examining impacts on multiple outcomes. At the same time, exploratory findings must be considered with caution.

To test whether program impacts differed for subgroups of children, we estimated models with interaction terms that allowed the effect of experimental assignment to vary depending on child age at referral (13 or older versus younger than 13) and on type of site (the public agency versus the private agency sites). We did not examine subgroup impacts on outcomes assessed in the interview data due to the already-reduced sample sizes.

Findings

Children who were assigned to the intervention services were no more likely than control group children to experience a positive move during the study period. Findings for the confirmatory outcome, as well as for the many exploratory outcomes, appear in Tables 3 (administrative data results) and 4 (interview data results). Results for alternative approaches to addressing missingness, clustering, and weighting appear in Online Appendix Tables 1 and 2.

Table 3.
Confirmatory and Exploratory Outcomes Assessed via Administrative Data, Regressed on Experimental Group Membership.

Outcomes n OR ^a B SE B

Confirmatory outcome

Any positive move (including discharge to permanency) 557 1.02 0.02 0.273

Exploratory outcomes

Components of the confirmatory outcome

Any positive foster care placement change 553 0.94 −0.06 .258

Any negative foster care placement change 553 1.37 0.32 .260

Compared with baseline foster care placement, placement at end of study was more positive 517 1.00 0.00 .294

Compared with baseline foster care placement, placement at end of study was more negative 517 1.26 0.23 .271

Permanency

Discharge to reunification^b 548 0.98 −0.02 .394

Discharge to adoption, guardianship, or reunification^b 564 0.88 −0.13 .246

Discharged from foster care 564 1.08 0.08 .295

Safety

Reallegation of abuse or neglect^b 542 0.90 −0.10 .261

Substantiated reallegation of abuse or neglect^b 537 0.36 −1.01 .497**

Placement restrictiveness

Discharged from foster care to a relative^b 558 0.91 −0.10 .324

Discharged from foster care to a relative OR discharged from foster care and last placement setting was with a relative 523 1.26 0.23 .376

Last placement setting in foster care is/was with a relative 523 1.18 0.17 .379

Placement stability

Reentry into care (among those discharged during study period)^b 349 0.90 −0.10 .534

Mean number of placements 564 1.15 0.14 .250

Note. Covariates include behavioral/emotional disability, any other disability, male gender, more than one spell of foster care, time in care at study enrollment, non-Hispanic White race, removed due to neglect, removed due to inadequate housing, placed in congregate care at baseline, placed in nonrelative foster home at baseline. Covariates for well-being models also include whether or not one or both of the youth’s parents are deceased. Missing data on covariates are addressed through mean imputation with indicators for missingness. OR = odds ratio. SE = standard error.

^aThe incident rate ratio from Poisson models, which is appropriate for count data, is reported for number of placements. ^bMultilevel model failed to converge. Results shown are from fixed-effects model with robust standard errors adjusted for clustering only at the sibling level.

p < .10. p < .05. p < .01.

Table 4.
Exploratory Outcomes Assessed via Interviews With Youth Age 13 or Older at Study Enrollment, Regressed on Experimental Group Membership.

Outcomes 12 months after enrollment 24 months after enrollment

N OR B SE B N OR B SE B

Behavior problems

Internalizing: clinical level 291 1.44 .37 0.368 265 1.16 0.47 .015

Externalizing: clinical level 294 1.88 .63 0.305 268 0.50 0.35 .150

Total problems: clinical level 291 1.63 .49 0.309 276 0.10 0.37 .793

Internalizing: borderline or clinical level 291 1.29 .26 0.271 268 0.26 0.33 .426

Externalizing: borderline or clinical level 292 1.14 .13 0.262 268 0.10 0.30 .741

Total problems: borderline or clinical level 291 1.22 .20 0.264 268 0.15 0.31 .614

Standardized social support scale score 279 NA .09 0.668 266 −0.47 0.83 .568

Self efficacy scale: range 18–72 303 NA −.80 1.062 278 −1.04 0.88 .240

Close with ≥1 child in household (among those in foster care living with other children) 149 1.83 .60 0.413 104 −0.20 0.70 .775

Close with any caregivers (in foster care only) 181 0.92 −.08 0.475 115 −0.65 1.01 .520

Contact with ≥1 sibling 246 0.57 −.56 0.367 213 −0.32 0.38 .409

Close with ≥1 sibling 286 1.54 .43 0.444 259 −0.11 0.41 .779

Assets

Has checking account 301 1.09 .09 0.321 279 −0.16 0.29 .583

Has savings account 299 1.53 .43 0.298 272 0.31 0.28 .268

Neither high school diploma nor general equivalency diploma (GED) 294 1.06 .06 0.443 280 0.19 0.35 .598

Enrolled in college 287 0.99 −.01 0.519 252 0.33 0.43 .445

Current living arrangement (multinomial model)^a

Nonkin foster home 290 0.50 −.68 0.322** 281 −0.57 0.34 .095

With kin or adoptive/biological parents (base)

Group home/residential 1.17 .16 0.401 0.04 0.41 .917

Independent, with roommate/significant other, other 1.19 .17 0.537 −0.08 0.39 .842

Contact with mother 263 1.09 .09 0.276 279 0.23 0.27 .391

Knows how to contact mother 254 0.98 −.02 0.294 279 0.36 0.31 .245

Close with mother 260 1.30 .27 0.272 279 0.15 0.27 .592

Contact with father 269 1.02 .02 0.284 278 −0.21 0.30 .487

Knows how to contact father 269 1.15 .14 0.265 278 −0.08 0.28 .772

Close with father 269 1.10 .09 0.310 278 −0.17 0.35 .632

Contact with ≥1 grandparent 302 1.31 .27 0.263 273 −0.07 0.29 .798

Close with ≥1 grandparent 302 1.48 .39 0.267 271 0.16 0.29 .579

Contact with ≥1 other relative 303 1.43 .36 0.263 279 0.03 0.30 .912

Contact with ≥1 other adult 298 0.98 −.02 0.257 268 −0.33 0.27 .230

Close with ≥1 other adult 295 0.96 −.04 0.255 274 −0.30 0.28 .272

Note. Data are weighted for survey nonresponse. Standard errors adjust for clustering by sibling groups, but not by counties. Covariates include behavioral/emotional disability, any other disability, male, ≥1 spell of foster care, time in care at study enrollment, non-Hispanic White race, removed due to neglect, removed due to inadequate housing, placed in congregate care at baseline, placed in nonrelative foster home at baseline, and whether or not one or both of the youth’s parents are deceased. Missing data on covariates were addressed through mean imputation with indicators for missingness. For the model predicting “current living arrangement” at the 12-month follow-up, single imputation was implemented only for whether or not the parents were deceased. OR = odds ratio. SE = standard error.

p < .10. p < .05. p < .01.

Overall, we tested differences in 71 outcomes between the groups; only 4 were statistically significant. The fact that about four significant differences should be expected due to chance alone should cause readers to suspect that even those results were due to chance. Keeping that caveat in mind, we found a lower odds for the treatment group than for the control group of living in nonrelative foster care rather than with kin or adoptive or biological parents, 12 months postbaseline. Second, treatment group members were more likely than control group members to have a clinical level of externalizing behaviors at the 12-month follow-up and to have a clinical level of internalizing behaviors at the 24-month follow-up. Lastly, the odds of a substantiated maltreatment allegation were lower for the treatment group than for the control group.⁶

The absence of an impact on the confirmatory outcome was robust regardless of (1) the method used for addressing missing data (single or multiple imputation), (2) the method for addressing the nesting of the data (multilevel model or fixed effect models with robust standard errors adjusting for sibling groups), and (3) whether or not county was controlled. The findings regarding behavior problems and maltreatment did not emerge consistently across analytical approaches. However, results from the unweighted, multiply imputed data were consistent with the potentially positive finding regarding the living arrangement at the 12-month follow-up.

We found no evidence that impacts on the confirmatory outcome—positive moves—vary by type of site or by age of child.⁷ We found one differential exploratory impact by type of site, with treatment group members in the counties served by a private agency experiencing more placement moves than control group members, but this finding was not robust across the modeling approaches tested.

Replication of Analyses of Impacts on Child Welfare and Safety Outcomes

We replicated analyses to the degree possible with data that had already been cleaned by the University of North Carolina’s School of Social Work. While we did identify some trivial differences, findings pertaining to impacts were consistent with the analyses we carried out using data we ourselves had cleaned. Overall, the replication provides some support that the findings are robust to different decisions about how the data should have been cleaned.

Summary and Discussion

The evaluation failed to find evidence that family finding has a positive impact, as indicated by positive moves to placement settings that were less restrictive, with relatives, and/or legally permanent. We also tested impacts on a large number of exploratory outcomes, finding four to be statistically significant. As a benchmark, if the outcomes were independent (which is unlikely to be the case), we would have expected impacts on four outcomes to emerge simply by chance. Exploratory evidence included possible favorable impacts on substantiated maltreatment allegations and on youths’ living arrangements 12 months following random assignment as well as potentially unfavorable impacts on behavior problems. No impacts emerged for the vast majority of the exploratory outcomes.

Limitations

Several limitations of the study are important to highlight, including a limited power to detect small effects due to our sample size. Also, we lacked a measure of emotional permanency. The interview asked youth about closeness and contact with relatives, but not about the expected permanency of these relationships. The follow-up time also presents a limitation; the administrative data allowed for a follow-up time ranging from 17 months to 4 years and 4 months, depending on the date youth were enrolled into the study. The survey allowed for 1- and 2-year follow-ups. Thus, we could not detect impacts that might have occurred over a longer time period.

We also experienced some challenges with data quality and missingness of data. Nonresponse and attrition are threats to the internal validity of findings based on the survey data. The use of weights may have reduced bias but was not likely to have eliminated it, given that the characteristics used to stratify the sample explained only a minority of the variability in nonresponse.

There are also limitations in the external validity. Counties excluded from the evaluation, which may have lacked capacity to implement the intervention, might differ systematically from counties participating in the evaluation. That said, the selected counties were diverse in terms of their geographic dispersion throughout the state as well as in the inclusion of both privately and publicly provided child welfare services and both rural and urban counties. In addition, the vast majority of the state’s children in out-of-home care reside in the selected counties.

Discussion

The analysis described in the present article study builds on an earlier analysis by Malm, Vandivere, Allen, Williams, and McKlindon (2014) by (1) carrying out imputation of missing data and (2) controlling for an expanded set of covariates theorized to affect the outcomes. The present study, using an ITT design, tested the effectiveness of the intervention as implemented in the field.⁸ Ultimately, however, we were unable to detect an impact on the confirmatory outcome.

One exploratory impact that we identified pertained to a favorable shift from nonrelative foster family homes to living with relatives. While this may at first seem incompatible with the lack of an impact on the confirmatory outcome of positive moves, the two measures differ. The confirmatory outcome came from administrative data on placement settings while in or immediately upon discharge from foster care. In contrast, the exploratory outcome came from youth reports about where they were living 12 and 24 months following random assignment, regardless of whether they were in foster care. Further, the exploratory outcome pertained to youth aged 13 and older at random assignment and may have been subject to nonresponse bias, whereas the confirmatory outcome came was available for all youth regardless of age.

Statistically significant impacts on four exploratory outcomes must be viewed with caution. Observed exploratory impacts may have occurred simply due to chance rather than as a result of the intervention. Further investigation and replication would be needed to determine the robustness of the exploratory findings and may also be warranted for the confirmatory outcome.

At the time the evaluations of family finding were first initiated, program developers expected the intervention to yield large positive changes, and evaluations were powered accordingly. Although some had more difficulty accumulating the planned sample sizes than others, the present study’s sample size (n = 568) is the second largest and is over twice the size of Iowa’s. The high expectations around family finding have probably also played a role in the widespread adoption of the model today. Yet the present study adds to a growing body of evidence, indicating that the expectations for family finding have not yet been realized. Any impacts occurring across evaluated family finding sites seem to be smaller than hoped, or perhaps are focused on emotional permanency, which was measured only in one evaluation.

Conclusions

Policy makers, administrators, and practitioners urgently need evidence regarding interventions for youth in foster care. Not only are youth in foster care a vulnerable population, but child welfare agency workers face many competing demands, so it is critical to know which efforts are most effective in various circumstances in achieving positive outcomes for children and youth. The findings here raise important future research and evaluation questions for family finding programs. In particular, can family finding be implemented in a way that results in consistently observed positive impacts? Or, are assumptions underlying the program model flawed? Does the intervention yield small or modest impacts, and/or are impacts apparent only after a longer period of observation? Does it affect emotional permanency? These are among the questions that should be the focus of the next generation of evaluations of family finding and other family search and engagement interventions.

Stages of study enrollment and data collection	Total	Treatment	Control
Underwent random assignment	573	295	278
Obtained child welfare administrative data	568	291	277
Eligible for interview (≥age 13 at study enrollment)	387	187	200
Obtained Round 1 interview data	305	143	162
Obtained Round 2 interview data	281	132	149

Child characteristics	Treatment (%)	Control (%)
Child disabilities (n = 567)
Emotionally disturbed	18	15
Other diagnosed condition	15	18
Child is male (n = 567)	58	57
Child age at study enrollment (n = 567)
Age 0–12	26	18*
Age 13–15	38	42
Age 16 or older	36	40
Child race and Hispanic origin (n = 561)
Non-Hispanic White	38	43
Non-Hispanic Black	53	47
Hispanic	5	7
Non-Hispanic other	4	3
More than one entry into foster care (including the initial removal during the study period; n = 567)	19	23
Time in foster care prior to study enrollment (n = 564)
Less than 1 year	18	24
At least 1 year but less than 2 years	27	18*
At least 2 years but less than 5 years	33	35
5 or more years	22	23
Type of foster care placement setting at referral (n =553)
In foster care, living with relatives	4	1*
Nonrelatives family foster home	62	68
Congregate care setting	34	31
Removal reason (for episode of care during study period; n = 567)
Relinquishment/abandonment	7	12
Physical abuse	10	10
Sexual abuse	7	6
Drug or alcohol abuse: child	1	2
Drug or alcohol abuse: parent	19	18
Child behavior problems	15	22
Child disability	1	3
Parent inability to cope	25	24
Parental death	4	3
Inadequate housing	8	14
Parents incarcerated	2	4
Neglect	77	70
Among youth age ≥13 at study enrollment^a (unweighted)
Status of parents (n = 304)
Parents deceased and/or unknown if either is living	14	6
One parent is known to be living	27	31
Both parents are known to be living	58	63
School/health professional ever told youth he/she had learning disability (n = 304)	31	22
Youth ever previously ran away (n = 304)	45	39

Outcomes	n	OR ^a	B	SE B
Confirmatory outcome
Any positive move (including discharge to permanency)	557	1.02	0.02	0.273
Exploratory outcomes
Components of the confirmatory outcome
Any positive foster care placement change	553	0.94	−0.06	.258
Any negative foster care placement change	553	1.37	0.32	.260
Compared with baseline foster care placement, placement at end of study was more positive	517	1.00	0.00	.294
Compared with baseline foster care placement, placement at end of study was more negative	517	1.26	0.23	.271
Permanency
Discharge to reunification^b	548	0.98	−0.02	.394
Discharge to adoption, guardianship, or reunification^b	564	0.88	−0.13	.246
Discharged from foster care	564	1.08	0.08	.295
Safety
Reallegation of abuse or neglect^b	542	0.90	−0.10	.261
Substantiated reallegation of abuse or neglect^b	537	0.36	−1.01	.497**
Placement restrictiveness
Discharged from foster care to a relative^b	558	0.91	−0.10	.324
Discharged from foster care to a relative OR discharged from foster care and last placement setting was with a relative	523	1.26	0.23	.376
Last placement setting in foster care is/was with a relative	523	1.18	0.17	.379
Placement stability
Reentry into care (among those discharged during study period)^b	349	0.90	−0.10	.534
Mean number of placements	564	1.15	0.14	.250

Outcomes	12 months after enrollment	24 months after enrollment
Behavior problems
Internalizing: clinical level	291	1.44	.37	0.368	265	1.16	0.47	.015**
Externalizing: clinical level	294	1.88	.63	0.305**	268	0.50	0.35	.150
Total problems: clinical level	291	1.63	.49	0.309	276	0.10	0.37	.793
Internalizing: borderline or clinical level	291	1.29	.26	0.271	268	0.26	0.33	.426
Externalizing: borderline or clinical level	292	1.14	.13	0.262	268	0.10	0.30	.741
Total problems: borderline or clinical level	291	1.22	.20	0.264	268	0.15	0.31	.614
Standardized social support scale score	279	NA	.09	0.668	266	−0.47	0.83	.568
Self efficacy scale: range 18–72	303	NA	−.80	1.062	278	−1.04	0.88	.240
Close with ≥1 child in household (among those in foster care living with other children)	149	1.83	.60	0.413	104	−0.20	0.70	.775
Close with any caregivers (in foster care only)	181	0.92	−.08	0.475	115	−0.65	1.01	.520
Contact with ≥1 sibling	246	0.57	−.56	0.367	213	−0.32	0.38	.409
Close with ≥1 sibling	286	1.54	.43	0.444	259	−0.11	0.41	.779
Assets
Has checking account	301	1.09	.09	0.321	279	−0.16	0.29	.583
Has savings account	299	1.53	.43	0.298	272	0.31	0.28	.268
Neither high school diploma nor general equivalency diploma (GED)	294	1.06	.06	0.443	280	0.19	0.35	.598
Enrolled in college	287	0.99	−.01	0.519	252	0.33	0.43	.445
Current living arrangement (multinomial model)^a
Nonkin foster home	290	0.50	−.68	0.322**	281	−0.57	0.34	.095
With kin or adoptive/biological parents (base)
Group home/residential		1.17	.16	0.401		0.04	0.41	.917
Independent, with roommate/significant other, other		1.19	.17	0.537		−0.08	0.39	.842
Contact with mother	263	1.09	.09	0.276	279	0.23	0.27	.391
Knows how to contact mother	254	0.98	−.02	0.294	279	0.36	0.31	.245
Close with mother	260	1.30	.27	0.272	279	0.15	0.27	.592
Contact with father	269	1.02	.02	0.284	278	−0.21	0.30	.487
Knows how to contact father	269	1.15	.14	0.265	278	−0.08	0.28	.772
Close with father	269	1.10	.09	0.310	278	−0.17	0.35	.632
Contact with ≥1 grandparent	302	1.31	.27	0.263	273	−0.07	0.29	.798
Close with ≥1 grandparent	302	1.48	.39	0.267	271	0.16	0.29	.579
Contact with ≥1 other relative	303	1.43	.36	0.263	279	0.03	0.30	.912
Contact with ≥1 other adult	298	0.98	−.02	0.257	268	−0.33	0.27	.230
Close with ≥1 other adult	295	0.96	−.04	0.255	274	−0.30	0.28	.272

Footnotes

Authors’ Note

Following the completion of the randomized controlled trial, which evaluated outcomes for clients served 2008–2011, and publication of initial findings in a Child Trends report (Vandivere et al., 2014), the private agency in North Carolina revised and expanded its relative search and engagement model and is no longer implementing family finding as described in the present article. The research protocol approved by our institutional review board prevents us from sharing any data pertaining to this study externally. However, readers may contact the lead author in order to ask any questions or to receive any materials that we can share, such as the interview questionnaire or tables with supplementary findings.

Acknowledgments

The authors thank Andrew Zinn of the University of Kansas School of Social Work, who provided advice and guidance on the quantitative analyses, and Mark Courtney of Chapin Hall at the University of Chicago, who provided helpful feedback on earlier drafts. In addition, we are grateful for the careful critiques provided by several anonymous reviewers, in addition to thoughtful comments from the editor, who synthesized the reviewers' feedback.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Generous funding from The Duke Endowment made this study possible.

Supplemental Material

Supplementary material for this article is available online.

Notes

References

Aarons

Sommerfeld

Walrath-Greene

(2009). Evidence-based practice implementation: The impact of public versus private sector organization type on organizational support, provider attitudes, and adoption of evidence-based practice. Implementation Science, 4, 83. doi:10.1186/1748-5908-4-83

Achenbach

T. M.

(1999). The child behavior checklist and related instruments. In Maruish

M. E.

(Ed.), The use of psychological testing for treatment planning and outcomes assessment (2nd ed., pp. 429–466). Mahwah, NJ: Lawrence Erlbaum.

Allison

P. D.

(2009). Missing data. In Millsap

R. E.

Maydeu-Olivares

(Eds.), The SAGE handbook of quantitative methods in psychology (pp. 72–89). Thousand Oaks, CA: Sage.

Andersson

(2005). Family relations, adjustment and well-being in a longitudinal study of children in care. Child and Family Social Work, 10, 43–56.

Avery

R. J.

(2010). An examination of theory and promising practice for achieving permanency for teens before they age out of foster care. Children and Youth Services Review, 32, 399–408.

California Permanency for Youth Project. (2008). Project evaluation. Retrieved from http://www.senecacenter.org/perm_evaluation

California Permanency for Youth Project. (2010). Project evaluation. Retrieved from http://www.senecacenter.org/perm_evaluation

Campbell

(2005). Six steps for family finding. San Leandro, CA: Center for Family Finding and Youth Connectedness. Resource Documents.

Campbell

(2010a). Basic family finding practice scaffold. Oakland, CA: National Institute for Permanent Family Connectedness, Seneca Center. Resource Documents.

10.

Campbell

(2010b). Family finding quality and fidelity index. Oakland, CA: National Institute for Permanent Family Connectedness, Seneca Center. Resource Documents.

11.

Collins-Camargo

McBeath

Ensign

(2011). Privatization and performance-based contracting in child welfare: Recent trends and implications for social service administrators. Administration in Social Work, 35, 494–516.

12.

Cook-Fuong

S. K.

(2000). The adult well-being of individuals reared in family foster care placements. Child and Youth Care Forum, 29, 7–25.

13.

Courtney

M. E.

Dworsky

(2006). Early outcomes for young adults transitioning from out-of-home care in the USA. Child & Family Social Work, 11, 209–219.

14.

Courtney

M. E.

Dworsky

Brown

Cary

Love

Vorhies

(2011). Midwest evaluation of the adult functioning of former foster youth: Outcomes at age 26. Chicago, IL: Chapin Hall at the University of Chicago.

15.

Deke

Puma

(2013). Coping with missing data in randomized trials. Evaluation Technical Assistance Brief #3 for OAH and ACYF Teenage Pregnancy Prevention Grantees. Washington, DC: Office of Adolescent Health, U.S. Department of health and Human Services.

16.

Dworsky

Courtney

M. E.

(2000). Self-sufficiency of former foster youth in Wisconsin: Analysis of unemployment insurance wage data and public assistance data. Madison, WI: IRP.

17.

Egbewale

B. E.

(2015). Statistical issues in randomised controlled trials: A narrative synthesis. Asian Pacific Journal of Tropical Biomedicine, 5, 354–359.

18.

Furstenberg

F. F.

(2010). On a new schedule: Transitions to adulthood and family change. The Future of Children, 20, 67–87.

19.

Garwood

M. M.

Williams

S. C.

(2015). Differing effects of family finding service on permanency and family connectedness for children new to versus lingering in the foster care system. Journal of Public Child Welfare. doi:10.1080/15548732.2015.1008619

20.

Gelman

(2007). Struggles with survey weighting and regression modeling. Statistical Science, 22, 153–164. doi:10.1214/088342306000000691. Retrieved from http://projecteuclid.org/euclid.ss/1190905511

21.

George

Bilaver

Joo Lee

Needell

Brookhart

Jackman

(2002). Employment outcomes for youth aging out of foster care. Chicago, IL: Chapin Hall Center for Children at the University of Chicago.

22.

Hines

A. M.

(1997). Divorce-related transitions, adolescent development, and the role of the parent-child relationship: A review of the literature. Journal of Marriage and the Family, 59, 375–388.

23.

James Bell Associates. (2013). Family connection discretionary grants: 2009-Funded grantees cross-site evaluation report—Final. Retrieved from http://www.nrcpfc.org/grantees_public/2009/Fam%20Conn%202009%20Cross-Site%20Final%20Report%206-17-13.pdf

24.

Knol

M. J.

Groenwold

R. H. H.

Grobee

D. E.

(2012). P-values in baseline tables of randomised controlled trials are inappropriate but still common in high impact journals. European Journal of Preventive Cardiology, 19, 231–232.

25.

Landsman

Boel-Studt

Malone

(2014). Results from a family finding experiment. Children and Youth services Review, 36, 62–69.

26.

Leathers

S. J.

Falconnier

Spielfogel

J. E.

(2010). Predicting family reunification, adoption, and subsidized guardianship among adolescents in foster care. The American Journal of Orthopsychiatry, 80, 422–431. doi:10.1111/j.1939-0025.2010.01045.x

27.

Malm

Vandivere

Allen

Williams

S. C.

McKlindon

(2014). A rigorous evaluation of family finding in North Carolina (Executive Summary). Bethesda, MD: Child Trends. Retrieved from http://www.childtrends.org/wp-content/uploads/2014/03/2014-09aEvaluationFFNC.pdf

28.

Maza

(2009). A comparative examination of foster youth who did and did not achieve permanency. In Kerman

Freundlich

Maluccio

A. N.

(Eds.), Achieving permanence for older children and youth in foster care (pp. 32–39). New York City, NY: Columbia University Press.

29.

Montgomery

Donkoh

Underhill

(2006). Independent living programs for young people leaving the care system: The state of the evidence. Children and Youth Services Review, 28, 1435–1448.

30.

Pecora

P. J.

Kessler

R. C.

Williams

O’Brien

Downs

A. C.

English

… Holmes

(2005). Improving family foster care: Findings from the Northwest foster care alumni study. Seattle, WA: Casey Family Programs.

31.

Perry

B. L.

(2006). Understanding social network disruption: The case of youth in foster care. Social Problems, 53, 371–391.

32.

Puma

Olsen

R. B.

Bell

S. H.

Price

(2009). What to do when data are missing in group randomized controlled trials (Publication # NCEE 2009-0049). Washington, DC: Institute of Education Sciences, National Center for Educational Evaluation and Regional Assistance, U.S. Department of Education.

33.

Royston

(2005). Multiple imputation of missing values: Update. Stata Journal, 5, 188–201.

34.

Schochet

(2008). Technical methods report: Guidelines for multiple testing in impact evaluations (Publication number NCEE 2008-4018). Washington, DC: National Center for Educational Evaluation and Regional Assistance, Institute for Education Sciences, U.S. Department of Education.

35.

Swartz

T. T.

Kim

Uno

Mortimer

O’Brien

K. B.

(2011). Safety nets and scaffolds: Parental support in the transition to adulthood. Journal of Marriage and Family, 73, 414–429.

36.

U.S. Department of Health and Human Services. (2014). The AFCARS Report (No. 21): Preliminary FY 2013 Estimates as of July 2014. Retrieved from http://www.acf.hhs.gov/sites/default/files/cb/afcarsreport21.pdf

37.

Vandivere

Malm

K. A.

(2015). Family finding evaluations: A summary of recent findings (Research Brief #2015-01). Bethesda, MD: Child Trends. Retrieved from http://www.childtrends.org/wp-content/uploads/2015/01/2015-01Family_Finding_Eval_Summary.pdf

38.

Wakcher

(2010). San Bernardino County’s California Permanency for youth project evaluation. Retrieved from http://www.senecacenter.org/perm_evaluation

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.08 MB