Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/283110923

Demographics,policy,andfostercarerates;APredictiveAnalyticsApproach

ARTICLEinCHILDRENANDYOUTHSERVICESREVIEW·NOVEMBER2015

ImpactFactor:1.11·DOI:10.1016/j.childyouth.2015.09.009

READS

14

2AUTHORS,INCLUDING:

JesseRioRussell

NationalCouncilonCrimeandDelinquency

18PUBLICATIONS18CITATIONS

SEEPROFILE

Allin-textreferencesunderlinedinbluearelinkedtopublicationsonResearchGate,

lettingyouaccessandreadthemimmediately.

Availablefrom:JesseRioRussell

Retrievedon:26February2016

Children and Youth Services Review 58 (2015) 118–126

Contents lists available at ScienceDirect

Children and Youth Services Review

j ourna l homepage: www.e lsev ie r .com/ locate /ch i ldyouth

Demographics, policy, and foster care rates; A PredictiveAnalytics Approach

Jesse Russell a,⁎, Stephanie Macgill b

a National Council on Crime and Delinquency, Madison, WI, United Statesb Best Friends Animal Society, Kanab, UT, United States

⁎ Corresponding author at: 426 S. Yellowstone Dr., Mad

http://dx.doi.org/10.1016/j.childyouth.2015.09.0090190-7409/© 2015 Published by Elsevier Ltd.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 18 November 2014Received in revised form 13 September 2015Accepted 14 September 2015Available online 16 September 2015

Keywords:Child maltreatmentFoster careFoster care entryTime in careChild welfare policyDemographicsRisk factorsMaltreatment risk factorsClassification and regression treesCARTs

Individual, family, and community-level factors have been suggested as explanations of foster care entry ratesand average lengths of time that children remain in foster care. They do not, however, provide a sufficient expla-nation of the substantial geographical variation in entry rates and average lengths of stay across theUnited States.State-level child welfare policies and state-level socioeconomic variables may help explain these trends, but noempirical analysis to date has identified how policies and socioeconomic facts might interact in ways that canhelp account for the wide geographic differences.Traditional statistical methods andmuch prior research have been unable to identify the combinatorial and non-linear interaction effects of the many suggested factors. A data set of 104 state-level variables was constructed tohelp answer the question of what accounts for geographic differences in foster care entry rates and averagelengths of stay in foster care. A predictive analytics approach (classification and regression trees) was used tosort through all the potential explanatory variables, their interactions, and combinations. The results show thatstate cultural orientations and socioeconomic facts together best explain foster care entry rates. In contrast,childwelfare policy and practice differences together best explain average lengths of stay in foster care. Interven-tions aimed at goals relating to who goes into foster care and how many children go into foster care might bemost effective if they focus on culture and socioeconomic facts. Interventions aimed to change lengths of timein care, on the other hand, might be most effective if targeted at state child welfare policies and practices.

© 2015 Published by Elsevier Ltd.

1. Introduction

There is a gap in the research literature regarding how child welfarepolicy along with individual, family, and community-level risk factorsmight all work together to help explain geographic variations in fostercare entry rates and length of time in care across the United States.This paper provides a better understanding of the differences in fostercare experiences across the United States by exploring what factors incombination most efficiently predict foster care entry rates and whatfactors in combination are the most efficient predictors of the amountof time that children spend in foster care.

The extant literature on foster care experiencesmight be best under-stood in terms of competing and accumulating risks: individual, family,community, and cultural. So many variables have some potential bear-ing that traditional methodologies cannot account for them all. Thispaper presents results from a nonlinear, nonparametric predictive ana-lytics approach that allows for multiple combinatorial, nonlinear inter-actions. This type of predictive analytics modeling is the best way tosort through all factors in order to identify those that empirically matterthe most.

ison, WI 53719.

There is considerable geographic variance in foster care entry ratesand in how long children remain in care across states. For example, in2010, the foster care entry rate in Indiana was 5.7 per 1000 childrenand in North Carolina it was 2.1. Put another way, children in Indianawere nearly three times as likely as children in North Carolina to beplaced into foster care in 2010. The use of foster care as a state interven-tion was clearly different between those two states.

Foster care entry rates have a tendency to vary. For example, the fos-ter care entry rate increased from 4.7 per 1000 children in 1980 to 7.7per 1000 children in 2000 (Wertheimer, 2002). The foster care entryrate trend later reversed, falling to 3.6 per 1000 children in 2010 (U.S.Department of Health and Human Services, 2011).

Similarly, there is considerable variation in the average time thatchildren spend in foster care across the United States. In 2010, childrenremained in care an average of 630 days (20 months) once removedfrom the home, not counting the length of any previous foster care ep-isodes. This average is consistent with length-of-stay trends since the1990s (Barbell & Freundlich, 2001); however, how long children stayin foster care varies significantly from state to state. In Connecticut, chil-dren who exited foster care in 2010 had been in care an average of810 days (almost 27 months), while children in New Jersey who exitedin 2010 had been in care an average of 614 days (20 months). Children

119J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

in Illinois exiting care in 2010 had been in care an average of 1336 days(44 months; U.S. Department of Health and Human Services, 2010).

These inconsistencies across states in average lengths of stay in fos-ter care are troubling because there is no clear standard for the level ofappropriate state response to child maltreatment. As one analyst hassuggested, while more aggressive use of foster care might increasechild safety, removal from parents can be traumatic to children as well(Doyle, 2007). The wide differences across states in average lengths ofstay indicate that there is clear way to balance removing a child froman unsafe environment and the trauma experienced by children fromlingering in foster care. Not having a complete understanding of statevariations in how this balance is made can hinder interventions aimedat improving outcomes related to foster care entry and lengths of stay.

2. Potential explanations and variables

Potential explanations for these variations across states in foster carehave been suggested at the individual, the family, and the communitylevels.

On the individual level, children with special needs that increasecaregiver burden and children younger than four years of age face an in-creased risk of abuse or neglect (Centers for Disease Control andPrevention, 2012). A child's race may also be correlated with a greaterlikelihood of foster care placement. In almost every state, AfricanAmerican children are overrepresented in foster care when comparedwith the general population (Summers,Wood, & Russell, 2013). A num-ber of parental characteristics, such as age and level of education, alsoplace children at higher risk. Further, parental lack of understanding ofchildren's needs, child development and parenting skills, a parent'sown history of maltreatment, and parental substance abuse andmentalillness are all risk factors (Centers for Disease Control and Prevention,2012; Hines, Lemon, Wyatt, & Merdinger, 2004).

On the family level, many factors have been found to be potentiallyassociated with child maltreatment risk. The factors, of course, vary byfamily, but may include parental incarceration; low socioeconomic sta-tus; havingmore than four children; social isolation; family disorganiza-tion, dissolution, and violence; parenting stress; poor parent–childrelationships; and negative interpersonal interactions (Centers forDisease Control and Prevention, 2012; Sedlak et al., 2010; WorldHealth Organization, 2002). Parents who were themselves maltreatedas children are also at greater risk of maltreating their own children(Dixon, Browne, & Hamilton-Giachritsis, 2005).

On the community level, there are a number of factors that relate toincreased child maltreatment risk (Garbarino & Crouter, 1978). Lery(2009) found that elements of neighborhood social structure—in particu-lar, residential instability, impoverishment, and child care burden—wererelated to the risk of entry into foster care. Community violence, acuteneighborhood disadvantage (e.g., high poverty and high unemploymentrates), poor social capital, and high population density are also relatedto child maltreatment risk (World Health Organization, 2002).

Beyond community structure, culture or common communityvalues might relate to child maltreatment risk. For example, the preva-lence of collectivist values (interdependence, social support, and a “we”mindset) in a community versus prevalence of individualist values (in-dependence, self-reliance, and an “I”mindset) could shape responses tomaltreatment incidents (Vandello & Cohen, 1999). A critical aspect ofindividualism and collectivism for the purposes of this study is the dif-ferent emphasis on relationships in each construct. Individualists viewrelationships and groups as impermanent and non-intensive andthrough a cost–benefit lens. Collectivists, on the other hand, see rela-tionships and group memberships as permanent, stable, relatively im-permeable, and important (Oyserman, Coon, & Kemmelmeier, 2002).The role of culture has been examined within the child maltreatmentcontext, yet further understanding of its role is necessary (see, for exam-ple, Elliott & Urquiza, 2006; Stoltenborgh, Bakermans-Kranenburg, vanIjzendoorn, & Alink, 2013).

The individualism-collectivism index used here from Vandello andCohen (1999) is based on an extensive research history of culturalsocial-psychology constructs. These construct have be found to help ex-plain patterns in behaviors, cognition, attitudes, goals, values and familystructures (see Triandis, 1996). Vandello and Cohen summarize theconstruct as “collectivism can be defined as a social pattern of closelylinked individuals who define themselves interdependent members ofa collective (e.g., family and coworkers), whereas individualism as a cul-tural pattern stresses individual autonomy and individualism of theself” (p. 279). The index used is based on eight items: (1) percentageof people living alone (reverse scored); (2) percentage of elderly people(aged 65 and over) living alone (reverse scored); (3) percentage ofhouseholds with grandchildren living in them; (4) divorce to marriageratio (reverse scored); (5) percentage of people with no religious affili-ation (reverse scored); (6) average percentage number of people votingLibertarian over the last four presidential elections (reverse scored);(7) ratio of people carpooling to work to people driving alone; and(8) percentage of self-employed workers (reverse scored).

An alternative explanation for the differences in foster care experi-ences may be variations in child welfare policy across states, whichcan have an important effect on children's risk of maltreatment. Childwelfare policies vary from state to state, though all are built on the struc-ture provided through federal law. States differ primarily in how theyhave implemented laws and sometimes in laws themselves (in defini-tions of abuse and neglect, for example).

Federal laws (e.g., the Child Abuse Prevention and Treatment Act,the Adoption Assistance and Child Welfare Act, the Adoption and SafeFamilies Act, the Fostering Connections to Success and Increasing Adop-tions Act, the Indian Child Welfare Act, and the Multiethnic PlacementAct) provide an important statutory framework regarding how statesensure children's safety, permanency, and well-being. However, statesdiffer in the nature of implementation of these laws. For example, statesvary in terms of mandatory reporter laws and statutory definitions ofchild abuse and neglect, and not all have extended foster care to youthwho are past the age of 18.

3. Current study

The current study addresses the gap in literature regarding howchild welfare policy and risk factors work together to explain variationsfrom state to state in foster care entry rates and length of time in care.This study seeks to better understand the differences in children'schild welfare system experiences across the United States by exploringthe following research questions.

Research Question 1: What factors most efficiently predict fostercare entry rates across states?

Research Question 2: What factors most efficiently predict theamount of time children spend in foster care across states?

The extant literature on foster care experiencesmight be best under-stood in terms of competing and accumulating risks: individual, family,community, and cultural. So many variables have some potential bear-ing that traditional methodologies cannot account for them all. A non-linear, nonparametric model that allows for multiple combinatorial,nonlinear interactions is the best way at this stage of the research pro-gram to sort through all factors in order to assess those that matterthe most.

4. Method

4.1. Participant characteristics

The individual states in theUnited Stateswere selected as the unit ofanalysis. Themost recent data up to 2010were usedwhen possible. TheDistrict of Columbia was excluded from the data set because it was an

Fig. 1. Collectivismwas the singlemost efficient predictor of foster care entry rates amongstates.

120 J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

outlier along several variables, such as average income and averagechild age, and because it was missing data for a number of variables.

4.2. Data set

State-level data on child welfare policies, outcomes, and demo-graphics were collected from 10 sources: the ChildWelfare InformationGateway State Statutes Search (U.S. Department of Health and HumanServices, nda); the Child Welfare Enacted Legislation Database(National Conference of State Legislatures, 2012); the KIDS COUNTData Center (Annie E. Casey Foundation, nd); the Reports and Resultsof the Child and Family Service Reviews (U.S. Department of Healthand Human Services, ndb); the Fostering Connections Legislation byState Database (Fostering Connections Resource Center, nd); the StateChildWelfare Policy Database (Casey Family Programs, 2010); theWel-fare Rules Database (The Urban Institute, 2011); Census 2010 (U.S.Census Bureau, nd); and the Adoption and Foster Care Analysis andReporting System (AFCARS) (U.S. Department of Health and HumanServices, 2010). Social values for each state were taken from Vandelloand Cohen (1999).

4.3. Measures

4.3.1. Foster care entry rateThe foster care entry rate was calculated using 2010 AFCARS and

2010 census data. AFCARS is a federal reporting system maintained byDHHS that collects case-level information on all children in foster care.In the AFCARS database, each child is counted only once and the infor-mation included on the child in the database is from the child's most re-cent foster care episode (U.S. Department of Health and HumanServices, 2012; U.S. Department of Health and Human Services, 2010).For each state, the total number of foster care entries was divided bythe total child population and multiplied by 1000 to create a fostercare entry rate per 1000 children in each state.

4.3.2. Average length of time in careAverage length of time in foster care was derived from 2010 AFCARS

data by calculating the length of a child's stay in foster care from thedate of their latest removal to their exit date. The start of the current ep-isode could be in any year, and the exit was in fiscal year 2010.

4.3.3. Independent variablesA total of 104 independent variables were entered into a database,

relating to legal definitions of abuse, neglect, and domestic violence;statutes regardingmandatory reporting, kinship care, foster care licens-ing, and parent representation; degrees of implementation of differen-tial response systems and Fostering Connections; Child and FamilyServices Review (CFSR) ratings; poverty indicators; U.S. Census Bureaudemographics; AFCARS data; Temporary Assistance to Needy Families(TANF) rules; or social values. Some variables of note are explained inmore detail below. See Appendix A for a detailed presentation of inde-pendent variables and their sources.

4.4. Analysis

Classification and regression trees (CARTs) are a form of predictiveanalytics using a recursive partitioning algorithm. CARTs attempt to ac-curately and reliably predict an outcome of a particular case based onthe characteristics of that case (Kass, 1980; Breiman, Friedman,Olshen, & Stone, 1984). The CART algorithm searches for patterns(which may be nonlinear) and combinatorial relationships in a set ofdata. CARTs are particularlywell suited for uncovering hidden nonlinearstructures and interactions in complex data sets. Further, CARTs are anappropriate method for exploratory analysis where the number of pre-dictor variables is greater than the sample size (Strobl, Malley, & Tutz,2009) or when there is not a theoretical basis for hypothesizing how

predictor variablesmight interact. CARTs for the current studywere cre-ated using the open source R statistics and programming platform. Theparticular package used was the RPART routine, which is based on thespecifications of Therneau and Atkinson (1997a, 1997b). While this sta-tistical approach is not often found in social science scholarship, it is anideal method for the current study. The advantage of the approach isthat it reveals the contingent relationships that help to determine thedependent variables. In addition, the approach can tell us which vari-ables are the most efficient predictors (Russell, 2011).

CARTs are created with an algorithm that seeks to maximize homo-geneity across values of the dependent variable within resulting treenodes. For each point on the scale of each variable, the algorithm testshowwell that particular point on that particular variable could separateall the observations in the data set into two more homogenous groups.Once every point on every variable is tested, the algorithm selects theone that best separates all the observations into homogenous groups.Then, for each of these two new groups, the algorithm proceeds againto identify which point on which variable could separate the groupinto two further more homogenous subgroups.

The algorithm seeks a balance between the most homogenousgroups possible and over-fitting themodel. This is achieved bymeasur-ing additional model complexity (such as the number of splits) as atradeoff with additional homogeneity with each potential split (mea-sured with something like the Akaike information criterion). Balanceis also achieved by dividing the full data set into randomly drawn sub-samples, developing multiple models, and cross-validating them witheach other. In the end, the model can be judged according to how wellit parsimoniously and accurately classifies all the observations in thedata set.

Fig. 2. Child welfare expenditures was the singlemost efficient predictor of length of timein care rates among states.

121J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

5. Results

The results are presented in the trees in Figs. 1 and 2. It is importantto note that while the recursive partitioning algorithm sequentiallymakes computations on all the explanatory variables and calculatesevery possible split for each variable, the final results rely on splitsmade on only a few variables. The variables and splits represented inFigs. 1 and 2 are themost efficient in their ability to predict average fos-ter care entry rates and average length of time in care.

The tree diagrams (shown in Figs. 1 and 2) can be read from the topdown. The full sample of cases is present at the top. The sample is thensplit according to a sequence of criteria. At each split a single criterion isgiven. The criterion is based on one independent variable, with highervalues going oneway and lower values going anotherway. For example,a split criterion might be based on a particular variable having valuesgreater or equal to 14.5. Cases for which the variable has a value of 15or greater would be split to the left, and cases for which the variablehas a value of 14 or less would be split to the right. This process of splitscontinues down the tree.

At the bottom of the tree are final nodes, which are not split any fur-ther. The number below the node title is the average foster care entryrate or average length of stay in care for the states in that node. The nsize indicates how many states are in that particular node along withthe percentage of observations in that node (out of 50).

5.1. Foster care entry rate

Model 1 shows how the CART algorithm separated all 50 observa-tions through a series of binary splits to create groups (nodes) with

similar foster care entry rates. The CART algorithm identified collectiv-ism as the most effective primary split (Fig. 1). Vandello and Cohen(1999) developed a collectivism index that ranked each state's collectiv-ism score; higher scores (on a scale of 1 to 100) indicatemore collectiv-ism. In the first split, states with collectivism scores lower than 40moveto the right to Node 1. The average foster care entry rate for the states inNode 1 is 5.7 per 1000.

State diversity score was identified as the second split in Model 1.Each state's race data (the number of individuals in a particular racialor ethnic group), taken from the 2010 census, were entered into an on-line Shannon–Wiener Diversity Index tool (Chang Bioscience, 2011).The Shannon–Wiener Diversity Index is a term used in biology studiesandmeasures the rarity and commonness of species in a biological com-munity with the formula H = - ∑[pi * ln(pi)] -pi (pi). The online toolgenerated a score reflecting the state's level of diversity; higher scoresreflect greater diversity. States with diversity scores less than .87move to the right to Node 2 and have an average foster care entry rateof 4.2.

States with higher diversity scores were then split into a third groupaccording to the percentage of children living in crowded housing.Taken from the KIDS COUNTData Center, this variable is the percentageof children under age 18 living in households that havemore than 1 per-son per room. States with 12.5% or more of children living in crowdedhousing move to the right to Node 3 and have an average foster careentry rate of 3.7, while states with less than 12.5% move to the left toNode 4 and have an average foster care entry rate of 2.4.

The predicted foster care entry rates were compared with the actualfoster care entry rates (Table 1). Doing so demonstrates whether a stateis consistent with the other states in the same node or is an aberration.While there are several states in Nodes 1 and 2 whose actual foster careentry rates differ from the group average by over two points, there areno notable outliers. There is a smaller range of foster care entry ratesin Nodes 3 and 4.

To examine the predictive validity of the trees, we examined theamount of foster care reentry for each node. The average percentageof children who had been in foster care prior to 2010 was calculatedfor each node in each model and compared with each node's averagefoster care entry rates. In Model 1, the average percentage of reentriesfor Node 1was 26.3%. In otherwords, an average of 26.3%of the childrenliving in the states in Node 1 had been in foster care before their currentepisode. In Node 2, the average percentage of reentries was 22.9%. InNodes 3 and 4, the percentage of reentries was 19.8% and 18.7%, respec-tively. Overall, the highest reentry averagesmatched the nodeswith thehighest entry rates. A Pearson correlation coefficient was calculated toassess the relationship between foster care entry rates and levels of re-entry. There was a positive correlation between the two variables (r =.41, p = b .01), indicating a moderately strong relationship.

5.2. Time in care

Model 2 presents how all 50 observations can be divided through aseries of binary splits into groups (nodes) with similar length of timein care values. In Model 2, the CART algorithm identified a state's childwelfare expenditures as the most effective ways to split the full sampleinto higher and lower time in care values. Childwelfare expenditures in-clude the administration and operation ofmaltreatment prevention ser-vices for children and families, family preservation services, childprotective services, in-home services, out-of-home placements, andadoption services (Casey Family Programs, 2010). Stateswith childwel-fare expenditures of $551 million or greater move to the right (seeFig. 2).

States with lower child welfare expenditures were split according tothe minimum age at which youth are eligible for supervised indepen-dent living. Each state has its own standard for the appropriate age atwhich youth can becomeeligible for supervised independent living pro-grams. This variable represents only the age at which youth may begin

Table 1Model 1 results compared to actual values for each state.

State Model 1node

Node average foster careentry rate

Actual foster careentry rate

Washington 1 5.7 4Montana 1 5.7 4.3Colorado 1 5.7 4.7Kansas 1 5.7 4.9North Dakota 1 5.7 5.5Oregon 1 5.7 5.6Iowa 1 5.7 6.5South Dakota 1 5.7 7.2Wyoming 1 5.7 7.3Nebraska 1 5.7 7.4New Hampshire 2 4.2 1.8Utah 2 4.2 2.6Maine 2 4.2 2.8Idaho 2 4.2 3.2Ohio 2 4.2 3.4Wisconsin 2 4.2 3.4Michigan 2 4.2 3.5Pennsylvania 2 4.2 3.6Missouri 2 4.2 4.2Tennessee 2 4.2 4.2Vermont 2 4.2 4.3Minnesota 2 4.2 4.4Kentucky 2 4.2 5.4Arkansas 2 4.2 5.6Indiana 2 4.2 5.7Rhode Island 2 4.2 6.5West Virginia 2 4.2 7.6Texas 3 3.7 2.4New York 3 3.7 2.9Louisiana 3 3.7 3Hawaii 3 3.7 3.4New Mexico 3 3.7 3.4Mississippi 3 3.7 3.4California 3 3.7 3.6Florida 3 3.7 3.6Nevada 3 3.7 4.3Oklahoma 3 3.7 4.7Alaska 3 3.7 4.8Arizona 3 3.7 4.8Virginia 4 2.4 1.5Illinois 4 2.4 1.7Delaware 4 2.4 2Maryland 4 2.4 2.1North Carolina 4 2.4 2.1Georgia 4 2.4 2.2New Jersey 4 2.4 2.3Alabama 4 2.4 2.7Connecticut 4 2.4 3South Carolina 4 2.4 3.1Massachusetts 4 2.4 3.9

Table 2Model 1 results compared to actual values for each state.

State Model 2node

Node average length oftime in care (days)

Actual average length oftime in care (days)

Indiana 1 761 486Ohio 1 761 560Florida 1 761 562New Jersey 1 761 614Pennsylvania 1 761 619Massachusetts 1 761 626Washington 1 761 659California 1 761 769Texas 1 761 773Virginia 1 761 799Connecticut 1 761 810Michigan 1 761 838New York 1 761 873Maryland 1 761 1095Illinois 1 761 1336Arkansas 2 521 354Minnesota 2 521 377Wyoming 2 521 377Tennessee 2 521 413New Mexico 2 521 421Colorado 2 521 468Utah 2 521 473South Carolina 2 521 497Louisiana 2 521 534Iowa 2 521 551West Virginia 2 521 559Delaware 2 521 585Missouri 2 521 683Oregon 2 521 745Oklahoma 2 521 784South Dakota 3 525 359North Dakota 3 525 472Idaho 3 525 479Kentucky 3 525 483Hawaii 3 525 523Mississippi 3 525 528Rhode Island 3 525 563Kansas 3 525 643Alabama 3 525 680Arizona 4 686 484Georgia 4 686 587Wisconsin 4 686 591Nebraska 4 686 611Nevada 4 686 647Vermont 4 686 685North Carolina 4 686 694Alaska 4 686 771Montana 4 686 776New Hampshire 4 686 804Maine 4 686 896

122 J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

participating in supervised independent living programs. States with el-igibility ages 16 or younger move to the left to Node 2 and have an av-erage length of stay in care of 521 days.

States thatmoved to the right in terms of supervised independent liv-ing eligibilitywere split according to their scores on CFSR Safety Outcome1, that children are first and foremost protected from abuse and neglect:“Of all children who were victims of substantiated or indicated childabuse and/or neglect during the first six months of the year, what per-centage had another substantiated or indicated report within a six-month period?” (U.S. Department of Health and Human Services, 2013).Mandated by ASFA, the CFSRs are one of the ways in which the federalgovernment exercises oversight of the child welfare system. Evaluatorsassess state performance along a broad range of systemic, family, andchild outcome measures to determine how well states are performing(Allen & Bissell, 2004; U.S. Department of Health and Human Services,2011). Safety Outcome 1 reflects the recurrence of substantiated or

indicated maltreatment. The CFSR ranks each state's achievement of thisoutcome as a percentage (0 to 100). States with scores greater than orequal to 84% (e.g., higher conformity) on Safety Outcome 1 move to theleft to Node 3 and have average lengths of stay in care of 525 days. Stateswith less conformity move to the right to Node 4 and have averagelengths of stay of 686days. AswithModel 1, the node averageswere com-pared with the actual state averages for Model 2 (Table 2). There arestates in each node whose average length of stay in care is greater orless than the group average by as many as 200 days.

Foster care reentries were also examined for Model 2. In Node 1, theaverage percentage of children who had reentered the system was21.1%. In Node 2, reentry was 22.7%. Nodes 3 and 4 averaged 22.7%and 21.2%, respectively. Reentry levels across nodeswere, therefore, vir-tually equal. How long children remain in care does not appear to be re-lated to the degree to which children reenter care. If reentries weresystematically related to how long children remain in care, we would

123J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

expect to see a more linear relationship between reentry volume andaverage lengths of stay in care. Furthermore, the Pearson coefficient in-dicated no relationship between the two variables (r=−.12, p= .41).

6. Discussion

The purpose of this study is to begin to better understand state var-iations in terms of foster care entry rates and how long children remainin foster care. To that end, a data set with more than 100 child welfarepolicy-related and demographic variables was constructed. The CARTalgorithms revealed that different factors most efficiently predict fostercare entry rates and average lengths of stay in care.

In terms of foster care entry rates, a state's score along the socialvalue spectrum was the primary split predicting foster care entryrates. States that scored lower on the collectivism scale or higher onthe individualism scale have higher average foster care entry rates. Inother words, states with the most individualistic orientation removechildren with more frequency than collectivistic states. This can be un-derstood when examining the differences between an individualisticculture's emphasis on independence, self-reliance, and personal dispo-sitions and a collectivistic culture's emphasis on interdependence, socialsupport, and situational attributes. Individualistic cultures are more aptto consider the individual when something happens, while collectivisticcultures are more apt to examine the role situational factors play in cri-ses. For example, in an individualistic culture, a mother who has a drugproblem is personally responsible for those decisions and needs to haveher children removed so that she can help herself. In a collectivistic cul-ture, a mother with a drug problem may use drugs because she has nojob, skills, or support system and removing the children may not help,but appropriate services might.

Less-diverse states have the second highest average foster care entryrates. The results indicate that states with smaller minority populationsremove children at higher rates thanmore diverse states. This finding isin accord with Foster (2012), who found that states with larger AfricanAmerican populations are less likely to take children into protective cus-tody, and who points to the need for a better understanding of the roleof race in child welfare decision-making.

Finally, foster care entry rates in states that are bothmore collectivistand more diverse are predicted by how many children live in crowdedhousing. States with more children living in crowded housing havehigher entry rates than states with fewer children in crowded housing.Understanding this variable as a measure of poverty helps explain thisfinding and corroborates explanations of poverty driving foster careentry.

Not only do foster care entry rates appear to be explained by a state'scollectivistic orientation, diversity, and level of poverty, but entry ratesare alsomoderately correlatedwith a state's level of reentry, suggestingthat these variables can also help bolster understanding as to why chil-dren reenter foster care. It could be that families in individualistic statesare not receiving enough services to prevent maltreatment recurrencebecause the individualistic orientation places emphasis on individualresponsibility over the intensive social services often needed to addressthe causes of child maltreatment (e.g., mental health issues andsubstance abuse). Secondly, families in individualistic states may not

Variable

1 Foster care entry rate

Appendix A. Variables and sources

be receiving the same type of help in their services. For example, indi-vidualistic services may be more focused on changing the individual,while services in collectivistic contexts may bemore focused on chang-ing an individual's surroundings—to say nothing of a state's fiscal con-text and ability to maintain a comprehensive menu of social services.

Average lengths of time in care in states with lower childwelfare ex-penditures were then split according to the minimum age at which fos-ter youth are eligible for supervised independent living programs,which are designed to provide youth with the necessary skills for suc-cessful adulthood. States whose eligibility age is 16 or younger havethe lowest average foster care stays.

States that do not allow youth 16 or younger to enter supervised in-dependent living programs were then split according to their perfor-mance along CFSR Safety Outcome 1: that children are first andforemost protected from abuse and neglect. States with better perfor-mance along this measure had the second-lowest average length ofstay in care: four days longer than states that allow 16-year-olds toenter supervised independent living programs. On the other hand,states with less conformity with this measure had an average lengthof stay that was 161 days longer.

Conversely to Model 1, which explored foster care entry rates andfound demographic and cultural values to be most salient, the CART al-gorithm in Model 2 identified policy and child welfare system perfor-mance factors as the most efficient predictors. It would appear thatpolicy decisions do have consequences for how children experiencethe child welfare system.

The question of who goes into foster care—which children and howmany children—might be best understood in terms of factors relating tothe state broadly: collectivism, diversity, and housing trends. The “who”question appears to be less about policy decisions or practice ap-proaches. In contrast, questions about the foster care experience, partic-ularly lengths of time in care, might be best understood from policydecisions and practice orientations: expenditures, safety goals, andage-dependent programs.

The implications of this differencemight be first that it is difficult forpolicy and practice decisions to overcome underlying cultural and so-cioeconomic facts when pursuing goals related to who goes into fostercare or how many children go into foster care. Instead, these goalsmight be best approached by targeting cultural understanding and so-cioeconomic progress. Second, the implication might be that lengthsof stay in foster care can be understood as an attribute of how child wel-fare systems operate—their decisions, policies, and practice approaches.Goals related to safely reducing the amount of time children remain infoster care, for example, might be best approached through effectivepractice.

7. Conclusion

This paper explored what factors most efficiently predict foster careentry rates and how long maltreated children tend to remain in fostercare. A variety of variables were shown tomatter themost (communityvalues, demographics, poverty, expenditure levels, policy on olderyouth, and performance measures), and foster care reentry volumewas correlated with foster care entry rates.

Source

2010 Adoption and Foster Care Analysis Reporting System

(continued on next page)

(continued)

Variable Source

(AFCARS),2010 Census (http://factfinder2.census.gov)

2 Average length of stay in foster care 2010 Adoption and Foster Care Analysis Reporting System(AFCARS)

3 Total child welfare expenditures, FY 2010 Casey Family Programs, State Child Welfare Policy Database4 Percent change in total expenditures from FY 2008 Casey Family Programs, State Child Welfare Policy Database5 Definition of physical abuse includes acts/circumstances threatening child with harm

or creating substantial risk of harmCasey Family Programs, State Child Welfare Policy Database

6 Definition of neglect includes failure to educate child Casey Family Programs, State Child Welfare Policy Database7 Abandonment included in abuse or neglect status definition Casey Family Programs, State Child Welfare Policy Database8 Medical neglect defined in statute Casey Family Programs, State Child Welfare Policy Database9 Definition of abuse/neglect includes emotional maltreatment Casey Family Programs, State Child Welfare Policy Database10 Neglect depth Calculated from variables 4–711 State has specific definition of sex abuse Casey Family Programs, State Child Welfare Policy Database12 Specific definition of emotional abuse/mental injury Casey Family Programs, State Child Welfare Policy Database13 Parental substance abuse Casey Family Programs, State Child Welfare Policy Database14 Domestic violence defined in civil statutes Casey Family Programs, State Child Welfare Policy Database15 Civil statutes list specific acts constituting domestic violence Casey Family Programs, State Child Welfare Policy Database16 Domestic violence included in civil definition of child abuse Casey Family Programs, State Child Welfare Policy Database17 Domestic violence defined in criminal or penal code Casey Family Programs, State Child Welfare Policy Database18 Criminal statutes list specific acts constituting domestic violence Casey Family Programs, State Child Welfare Policy Database19 Number of exceptions Casey Family Programs, State Child Welfare Policy Database20 Status of differential response implementation Casey Family Programs, State Child Welfare Policy Database21 Number of distinct pathways/tracks for screened-in reports Casey Family Programs, State Child Welfare Policy Database22 State has an approved Title IV-E Guardianship Assistance Program plan under

Fostering ConnectionsCasey Family Programs, State Child Welfare Policy Database

23 State has an approved Title IV-E plan to extend care to older youth under Fostering Connections Casey Family Programs, State Child Welfare Policy Database24 Does state have current Fostering Connections legislation? Fostering Connections Resource Center, fosteringconnections.

org25 Number of Fostering Connections bills passed Fostering Connections Resource Center, fosteringconnections.

org26 Does the state have pending Fostering Connections legislation? Fostering Connections Resource Center, fosteringconnections.

org27 State uses narrow definition of kin as relatives (related by blood, marriage, or

adoption) being treated differently than other kin (godparents, etc.)Casey Family Programs, State Child Welfare Policy Database

28 State uses broad definition of kin defined as relatives and other kin having the sametreatment in all engagement with the child welfare agency

Casey Family Programs, State Child Welfare Policy Database

29 Number of methods used by child welfare agency to locate kin Casey Family Programs, State Child Welfare Policy Database30 State has ongoing involvement with kinship caregivers in private kin arrangements Casey Family Programs, State Child Welfare Policy Database31 State allows kin to be used as placement options to divert from foster care Casey Family Programs, State Child Welfare Policy Database32 State has a pre-approval process to allow children to be placed into kinship homes

almost immediately after they are removedCasey Family Programs, State Child Welfare Policy Database

33 Licensure options for kin caring for a child in state custody Casey Family Programs, State Child Welfare Policy Database34 State allows kin to pursue permanent legal guardianship Casey Family Programs, State Child Welfare Policy Database35 Guardianship payment compared to foster care payment Casey Family Programs, State Child Welfare Policy Database36 State policy gives preference to noncustodial parents over other relatives Casey Family Programs, State Child Welfare Policy Database37 State statutory descriptions of who is required to report suspected or known child

abuse or neglectCasey Family Programs, State Child Welfare Policy Database

38 Other professionals who are specified in state statutes as mandatory reporters otherthan more commonly noted professionals (e.g., social workers, teachers, and physicians)

Casey Family Programs, State Child Welfare Policy Database

39 Youth can remain in foster care after 18th birthday under Fostering Connections Casey Family Programs, State Child Welfare Policy Database40 Youth can remain in foster care after 19th birthday under Fostering Connections Casey Family Programs, State Child Welfare Policy Database41 Cutoff foster care eligibility age Casey Family Programs, State Child Welfare Policy Database42 Number of circumstances allowing youth to stay in care past 18 (e.g., youth is on

track to graduate high school or get GED and youth has disabilities or special needs)Casey Family Programs, State Child Welfare Policy Database

43 State has requirements youth must comply with to remain in foster care after 18thbirthday (e.g., youth must be employed and youth must be enrolled in school)

Casey Family Programs, State Child Welfare Policy Database

44 Number of requirements youth must comply with to remain in foster care after 18th birthday Casey Family Programs, State Child Welfare Policy Database45 Number of circumstances allowing youth to stay in care past 19 (e.g., youth is on

track to graduate high school or get GED; youth has disabilities or special needs)Casey Family Programs, State Child Welfare Policy Database

46 State has requirements youth must comply with to remain in foster care after 19th birthday (e.g., youthmust be employed; youth must be enrolled in school)

Casey Family Programs, State Child Welfare Policy Database

47 Number of requirements youth must comply with to remain in foster care after 19th birthday Casey Family Programs, State Child Welfare Policy Database48 Court retains legal jurisdiction over youth in care after 18th birthday Casey Family Programs, State Child Welfare Policy Database49 Court retains legal jurisdiction over youth in care after 19th birthday Casey Family Programs, State Child Welfare Policy Database50 Supervised independent living is a placement option for youth in foster care Casey Family Programs, State Child Welfare Policy Database51 Minimum age for supervised independent living eligibility Casey Family Programs, State Child Welfare Policy Database52 Number of eligibility requirements for supervised independent living (e.g., youth

must be enrolled in school and youth must be working)Casey Family Programs, State Child Welfare Policy Database

53 Youth can reenter foster care after emancipation or discharge to independent living Casey Family Programs, State Child Welfare Policy Database54 Age at which foster youth become eligible for Chafee-funded services Casey Family Programs, State Child Welfare Policy Database55 State uses its own funds for independent living/transition services and supports Casey Family Programs, State Child Welfare Policy Database56 Number of state-funded services and supports (e.g., scholarships and housing subsidies) Casey Family Programs, State Child Welfare Policy Database57 Number of outcome items demonstrating high performance (rating of strength) on the Child and Family

Services Review (CFSR) second roundChildren's Bureau, Administration for Children and Families

58 CFSR Safety Outcome 1: Children are first and foremost protected from abuse and neglect Children's Bureau, Administration for Children and Families

124 J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

(continued)

Variable Source

59 CFSR Safety Outcome 2: Children are safely maintained in their homes when possible and appropriate Children's Bureau, Administration for Children and Families60 CFSR Permanency Outcome 1: Children have permanency and stability in their living situations Children's Bureau, Administration for Children and Families61 CFSR Permanency Outcome 2: The continuity of family relationships and connections is preserved for

childrenChildren's Bureau, Administration for Children and Families

62 CFSR Well-Being Outcome 1: Families have enhanced capacity to provide for theirchildren's needs

Children's Bureau, Administration for Children and Families

63 CFSR Well-Being Outcome 2: Children receive appropriate services to meet theireducational needs

Children's Bureau, Administration for Children and Families

64 CFSR Well-Being Outcome 3: Children receive adequate services to meet their physicaland mental health needs

Children's Bureau, Administration for Children and Families

65 CFSR Number Systemic Factors in conformity (Information System, QualityAssurance System, Service Array, etc.)

Children's Bureau, Administration for Children and Families

66 Number of circumstances in which reasonable efforts are not required(e.g., instances when the parent's rights to another child have been terminated)

Casey Family Programs, State Child Welfare Policy Database

67 Number of child maltreatment victims in 2010 Child Welfare Outcomes Report, Children's Bureau,Administration for Children and Families

68 Percentage of children in poverty KIDS COUNT Data Center, datacenter.kidscount.org69 Percentage of total population in poverty KIDS COUNT Data Center, datacenter.kidscount.org70 Percentage of children in single-parent families KIDS COUNT Data Center, datacenter.kidscount.org71 Percentage of children affected by foreclosure from 2007 to 2009 KIDS COUNT Data Center, datacenter.kidscount.org72 Percentage of parents without health insurance KIDS COUNT Data Center, datacenter.kidscount.org73 Percentage of children living in crowded housing KIDS COUNT Data Center, datacenter.kidscount.org74 Average monthly foster care payment The Adoption and Foster Care Analysis and Reporting System

(AFCARS),Children's Bureau, Administration for Children and Families

75 Percentage of male children in foster care The Adoption and Foster Care Analysis and Reporting System(AFCARS),Children's Bureau, Administration for Children and Families

76 Percentage of female children in foster care The Adoption and Foster Care Analysis and Reporting System(AFCARS),Children's Bureau, Administration for Children and Families

77 Average child age at start of fiscal year (October 1) The Adoption and Foster Care Analysis and Reporting System(AFCARS),Children's Bureau, Administration for Children and Families

78 To become a foster parent, state requires training prior to licensure by law Child Welfare Information Gateway State Statutes Search79 State provides a specific course of training Child Welfare Information Gateway State Statutes Search80 Require other specialized training (e.g. CPR, first aid, and fire prevention) Child Welfare Information Gateway State Statutes Search81 Number of training hours required Child Welfare Information Gateway State Statutes Search82 Number of minimum standard safety requirements for foster homes (e.g., smoke

detector, carbon monoxide detector, and fire extinguisher)Child Welfare Information Gateway State Statutes Search

83 Home must have sufficient number of bedrooms so children of opposite sex do not share Child Welfare Information Gateway State Statutes Search84 Require a home inspection by the state health department or by a fire marshal Child Welfare Information Gateway State Statutes Search85 State requires results of recent health exams Child Welfare Information Gateway State Statutes Search86 Federal background check required for applicant Child Welfare Information Gateway State Statutes Search87 Child abuse and neglect registry check required for applicant Child Welfare Information Gateway State Statutes Search88 Require check of child abuse and neglect registries in any other state where applicant has lived in past five

yearsChild Welfare Information Gateway State Statutes Search

89 Fingerprints required for prospective foster parents Child Welfare Information Gateway State Statutes Search90 Background checks required for all household members, regardless of age Child Welfare Information Gateway State Statutes Search91 State prioritizes relatives for out-of-home placements Child Welfare Information Gateway State Statutes Search92 Relative may be issued a temporary license but must be able to meet all requirements for full licensure

after temporary license expiresChild Welfare Information Gateway State Statutes Search

93 Relatives must meet all regulations for licensure before a child can be placed in their care Child Welfare Information Gateway State Statutes Search94 Child may be placed without formal licensing Child Welfare Information Gateway State Statutes Search95 Average income 2010 American Community Survey, United States Census96 Median income 2010 American Community Survey, United States Census97 Diversity score Calculated with data from American Fact Finder, United

States Census Bureau98 Collectivism score Vandello, J. A., & Cohen, D. (1999). Patterns of individualism

and collectivismacross the United States. Journal of Personality and SocialPsychology, 77(2), 279–292.

99 Parent representation in dependency cases is a state due process right (affirmed by acourt decision)

Sankaran, V. (nd). A national survey on a parent's right tocounsel in termination ofparental rights and dependency cases. Retrieved from:http://www.law.umich.edu/centersandprograms/ccl/specialprojects/Documents/National%20Survey%20on%20a%20Parent's%20Right%20to%20Counsel.pdf

100 Parent representation in dependency cases statutorily required Sankaran, V. (nd). A national survey on a parent's right tocounsel in termination ofparental rights and dependency cases. Retrieved from:http://www.law.umich.edu/centersandprograms/ccl/specialprojects/Documents/National%20Survey%

(continued on next page)

125J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

(continued)

Variable Source

20on%20a%20Parent's%20Right%20to%20Counsel.pdf101 Parent representation is afforded in statute but is also discretionary Sankaran, V. (nd). A national survey on a parent's right to

126 J. Russell, S. Macgill / Children and Youth Services Review 58 (2015) 118–126

References

Allen, M., & Bissell, M. (2004). Safety and stability for foster children: The policy context.The Future of Children, 14(1), 48–73.

Annie E. Casey Foundation (nd). KIDS COUNT data center. Retrieved from http://www.datacenter.kidscount.org.

Barbell, K., & Freundlich, M. (2001). Foster care today. Washington, DC: Casey Family Pro-grams (pp. 1–7).

Breiman, L., Friedman, J. H., Olshen, R. A., & Stone, C. J. (1984). Classification and regressiontrees. Belmont, CA: Wadsworth Publishing.

Casey Family Programs (2010). The state child welfare policy database. Retrieved fromhttp://www.childwelfarepolicy.org/maps.

Centers for Disease Control and Prevention (2012). Child maltreatment: Risk and protec-tive factors. Retrieved from http://www.cdc.gov/violenceprevention/childmaltreatment/riskprotectivefactors.html.

Chang Bioscience (2011). Shannon–Wiener diversity index/Shannon entropy calculator.Retrieved from: http://www.changbioscience.com/genetics/shannon.htm.

Dixon, L., Browne, K., & Hamilton-Giachritsis, C. (2005). Risk factors of parents abused aschildren: A mediational analysis of the intergenerational continuity of child maltreat-ment (part I). Journal of Child Psychology and Psychiatry, 46, 47–57.

Doyle, J. J. (2007). Child protection and child outcomes: Measuring the effects of fostercare. American Economic Review, 97(5), 1583–1610.

Elliott, K., & Urquiza, A. (2006). Ethnicity, culture, and child maltreatment. Journal of SocialIssues, 62(4), 787–809.

Foster, C. H. (2012). Race and child welfare policy: State-level variations indisproportionality. Race and Social Problems, 4(2), 93–101.

Fostering Connections Resource Center (nd). Legislation by state.Garbarino, J., & Crouter, A. (1978). Defining the community context for parent–child rela-

tions: The correlates of child maltreatment. Child Development, 49, 604–616.Hines, A. M., Lemon, K., Wyatt, P., & Merdinger, J. (2004). Factors related to the dispropor-

tionate involvement of children of color in the child welfare system: A review andemerging themes. Children and Youth Services Review, 26(6), 507–527.

Kass, G. V. (1980). An exploratory technique for investigating large quantities of categor-ical data. Applied Statistics, 29(2), 119–127.

Lery, B. (2009). Neighborhood structure and foster care entry risk: The role of spatial scalein defining neighborhoods. Children and Youth Services Review, 31, 331–337.

National Conference of State Legislatures (2012). Child welfare enacted legislation data-base. Washington, D.C.: Author.

Oyserman, D., Coon, H. M., & Kemmelmeier, M. (2002). Rethinking individualism and col-lectivism: Evaluation of theoretical assumptions and meta-analyses. PsychologicalBulletin, 128(1), 3.

Russell, J. (2011). Hidden patterns in exchange rate regime choice. Empirical Economics,40(2), 425–449.

Sedlak, A. J., Mettenburg, J., Basena, M., Peta, I., McPherson, K., & Greene, A. (2010). Fourthnational incidence study of child abuse and neglect (NIS-4). Washington, DC: U.S. De-partment of Health and Human Services Retrieved from: http://www.law.harvard.edu/programs/about/cap/cap-conferences/rd-conference/rd-conference-papers/sedlaknis.pdf.

Stoltenborgh, M., Bakermans-Kranenburg, M. J., van Ijzendoorn, M. H., & Alink, L. R.(2013). Cultural–geographical differences in the occurrence of child physical abuse?

A meta-analysis of global prevalence. International Journal of Psychology, 48(2),81–94.

Strobl, C., Malley, J., & Tutz, G. (2009). An introduction to recursive partitioning. Munich,Germany: Department of Statistics, University of Munich.

Summers, A., Wood, S., & Russell, J. (2013). Disproportionality rates for children of color infoster care (Technical Assistance Bulletin). Reno, NV: National Council of Juvenile andFamily Court Judges.

The Urban Institute (2011). The welfare rules database.Washington, DC: Author Retrievedfrom http://anfdata.urban.org/wrd.

Therneau, T. M., & Atkinson, E. J. (1997a). An introduction to recursive partitioning usingthe RPART routines (Technical Report 61, Section of Biostatistics). Retrieved fromMayo Clinic website: http://www.mayo.edu/hsr/techrpt/61.pdf.

Therneau, T. M., & Atkinson, E. J. (1997b). An introduction to recursive partitioning using theRPART routines. Technical Report 61. Rochester, MN: Mayo Clinic (p. 452).

Triandis, H. C. (1996). The psychological measurement of cultural syndromes. AmericanPsychologist, 51(4), 407.

U.S. Census Bureau (nd). Census 2010. American Fact Finder. Retrieved from http://factfinder2.census.gov/faces/nav/jsf/pages/index.xhtml.

U.S. Department of Health and Human Services (2010). Adoption and foster care analysisand reporting system. Washington, D.C.: Administration for Children and Families,Children's Bureau.

U.S. Department of Health and Human Services (2011). Federal child and family servicesreviews: Aggregate report, round 2. Washington, D.C.: Administration for Childrenand Families, Children's Bureau Retrieved from http://www.acf.hhs.gov/sites/default/files/cb/fcfsr_report.pdf.

U.S. Department of Health and Human Services (2012). About AFCARS. Washington, D.C.:Administration for Children and Families, Children's Bureau Retrieved from http://www.acf.hhs.gov/programs/cb/resource/about-afcars.

U.S. Department of Health and Human Services (2013). Child welfare outcomes2008–2011: Report to Congress. Washington, D.C.: Administration for Children andFamilies, Children's Bureau Retrieved from http://www.acf.hhs.gov/programs/cb/resource/cwo-08-11.

U.S. Department of Health and Human Services (nda). State statutes search. Washington,D.C.: Administration for Children and Families, Children's Bureau, Child Welfare In-formation Gateway. Retrieved from https://www.childwelfare.gov/systemwide/laws_policies/state/.

U.S. Department of Health and Human Services (ndb). Reports and results of the Childand Family Service Reviews. Washington, D.C.: Administration for Children and Fam-ilies, Children's Bureau. Retrieved from http://basis.caliber.com/cwig/ws/cwmd/docs/cb_web/SearchForm.

Vandello, J. A., & Cohen, D. (1999). Patterns of individualism and collectivism across theUnited States. Journal of Personality and Social Psychology, 77(2), 279–292.

Wertheimer, R. F. (2002). Youth who “age out” of foster care: Troubled lives, troubling pros-pects (Child Trends Research Brief). Baltimore, MD: Annie E. Casey Foundation Re-trieved from http://www.aecf.org/resources/youth-who-age-out-of-foster-care-troubled-lives-troubling-prospects/.

World Health Organization (2002). World report on violence and health. Geneva,Switzerland: Author.