Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Jan 1.
Published in final edited form as: Child Youth Serv Rev. 2021 Nov 9;132:106283. doi: 10.1016/j.childyouth.2021.106283

Family outcomes in alternative response: A multilevel analysis of recurrence

Stacey L Shipe a,*, Mathew C Uretsky b, Terry, V Shaw c
PMCID: PMC8896770  NIHMSID: NIHMS1756266  PMID: 35250134

Abstract

Alternative response (AR) is preventative, family-centered, strengths-based approach within child protective services (CPS). When AR is offered it typically creates a two-track system where low- to moderate-risk families are not subjected to a traditional, fact-finding response that concludes with a determination of child abuse/neglect. One area that continues to concern child welfare administrators and researchers is recurrence, or when a family returns to CPS. Yet, it is unclear whether AR families have the same or different predictors of recurrence than TR families. Using a multilevel analytic approach, the present study followed 17,741 families in one mid-Atlantic state for 18-months post-response to determine what child, family, and county-level predicted a reported re-investigation and a substantiated re-investigation. We found few differences in predictors at the child and family level but found distinct differences at the county level for AR families. Recommendations are provided for policy, practice, and research, including a suggestion for further inquiry on what makes an optimal AR track.

Keywords: alternative response, recurrence, multilevel analysis

1. Introduction

There were an estimated 4.3 million reports of suspected child abuse or neglect in 2018. A little more than half of those reports (56%) required a full investigation (U.S. Department of Health & Human Services, 2020). Just over one in ten of the cases (14%) were assigned to receive an alternative response (AR; U.S. Department of Health & Human Services, 2020). AR, also known as Differential Response, Family Assessment Response, Dual Track, or Multitrack System, differs from a Child Protective Services (CPS) investigation in that AR typically only accepts low to moderate risk cases (i.e., neglect and low-level physical abuse, not sexual abuse), though there is jurisdictional variation. AR caseworkers work to develop family rapport and offer family-centered preventative services designed to mitigate risk factors for future maltreatment. Although a traditional response (TR) also strives to be family-centered, the primary focus for caseworkers is collecting evidence to determine if child abuse or neglect occurred (Merkel-Holguin et al., 2015; Waldfogel, 2008). Moreover, TR cases end with a determination of whether the maltreatment occurred whereas this specific outcome is not part of an AR case. The flexibility of AR is one reason it has gained traction in the field and is currently implemented in over 30 states, Washington, DC. and multiple tribal regions (NCSL, 2019). This flexibility, however, presents challenges for practitioners and researchers seeking to make meaningful cross-jurisdiction comparisons (Hughes et al., 2013; Piper et al., 2019).

The rate at which families re-enter to the system (i.e., recurrence), is not uniformly defined or measured, presenting a continued challenge for child welfare administrators. For example, Connell et al. (2007), using the term re-referral, found some jurisdictions had re-referral rates as high as 85% over a 10-year period, whereas Jonson-Reid et al., (2017), using the term re-reports, found wide variation among states with some reporting a low of 11.8% (Virginia) to a high of 46.1% (Michigan) within a three-year period. It is important to note AR was not necessarily developed to prevent families from returning to the system, rather it was developed in part to prevent families from entering CPS by offering low risk families a non-authoritative intervention (National Quality Improvement Center on Differential Response in Child Protective Services [NQIC-DR], 2011; Waldfogel, 2009). Despite the widespread use of AR and the ongoing issue of recurrence, there is a scarcity of research that has focused specifically on which factors bring AR families back into the system. This study aims to provide additional evidence on factors related to CPS recurrence by looking at both re-investigations and substantiated re-investigations among families receiving either an AR or TR.

1.1. Alternative Response

AR was developed over 30 years ago in response to child welfare professionals’ concerns of providing an adversarial approach for lower risk families with less serious allegations (NQIC-DR, 2011). They further believed these families did not require a formal child maltreatment investigation resulting in substantiation of allegations. Moreover, by providing only a TR approach, families in need of preventative services were overlooked or their cases were closed, thus increasing their chances of formally entering the system (Waldfogel, 2009). The result was the development of AR, or two track system, where lower risk families enhanced their ability to safely care for their children by receiving case management and individualized services. This approach also limited the system’s mandated protective authority (Delaye & Sinha, 2017; Hughes et al., 2013; Waldfogel, 2009).

To date, there have been over 50 evaluations and studies on different outcomes of AR, indicating its rise in popularity (Piper et al., 2019). Given that a primary goal of AR is child safety, multiple studies of AR have demonstrated that children are equally safe in an AR as they are in TR, and that AR has reduced the number of children who formally enter the child welfare system (Fluke et al., 2019; Loman & Siegel, 2004, 2015; Merkel-Holguin et al., 2005). Studies have also identified a cost savings for AR compared to TR and found that both families and caseworkers are satisfied with the AR approach (Loman & Siegel, 2015; Siegel & Loman, 2006; Winokur et al., 2014). In an updated review of issues in AR, Piper et al. (2019) suggested child safety is still compromised due to jurisdictional differences in how families are sorted by risk level; and because it is unclear whether AR families actually receive more services, it is difficult to conclude that receipt of an AR is the reason for the reduction of families entering or returning to the system. However, Merkel-Holguin and Bross (2015) emphasized that AR was never meant to fix all the concerns of CPS. Instead, AR is a complementary track to a system that is already under scrutiny. Furthermore, AR is intended to be flexible and to meet the needs of families, thus, variability between jurisdictions is expected. Despite some inconstancies most studies have found the implementation of AR has resulted in similar or increased child safety and a reduction of recurrence.

1.2. Predictors of Recurrence

Recurrence is among the key child welfare outcomes examined across studies (e.g., Casanueva et al., 2015; Fluke et al., 2008; Marshall et al., 2010), yet how it is operationalized and measured, including the length of the follow-up period, differs due to limitations of available data and a lack of consensus on definition (Fluke et al., 2008). The most current approach is to use re-report, or a new report of child maltreatment after an initial investigation was completed (Bae et al., 2010). Re-investigation or re-referral has also been used and is defined as when a new investigation occurs after a previous CPS was closed, regardless of the final determination of the initial investigation (Connell et al., 2007). There is also use of the Child and Family Services Review definition of recurrence, which is the substantiation of a subsequent investigation after a previous investigation was substantiated (Carnochan et al., 2013). Another difference among recurrence studies is how long a family is followed post initial investigation. Some studies followed children and families for a short duration (e.g., 6 months) whereas others followed families for multiple years (Connell et al., 2007). The nuances of these definitions and measurement periods are important because it can change the overall rate of return. For example, Zhang et al. (2013) found that most re-reports occur within 6-months but that a subset of cases take as long as 25 months to return. Bae et al. (2013), on the other hand, found rates of recurrence did not widely differ based on whether a family had a previous unsubstantiated report (21.5%) versus a substantiated report (22%). What has been consistent, regardless of definition or follow-up period, is the majority of families will return within six-months of initial report and level off after about two years (Fluke et al., 2008; Zhang et al., 2013). For purposes of this study, we use two definitions of recurrence: 1) re-investigation, and 2) a substantiated re-investigation. Both are further described below.

In practice, decisions about track assignment, service receipt, and risk of future maltreatment are generally assessed via risk assessments. These assessments, however, have not proven to be highly sensitive or specific of the indicators of future maltreatment (i.e., less than 50% accuracy), likely due to different approaches to assessments across jurisdictions (e.g., actuarial or consensus-based; Van der Put et al., 2017). In particular, actuarial risk assessments are typically developed based on prior CPS data (Cuccaro-Alamin et al., 2017); therefore, it should allow caseworkers to be more accurate in assessing family need, including targeted services, so future CPS entry is mitigated (Coohey et al., 2013). Yet, low-risk families are regularly misclassified and do not receive the needed services leading to re-maltreatment and high-risk families are instead offered services and still re-maltreat; thus, returning to the system (Benbenishty et al., 2015; Cross & Casanueva, 2009; Detlaff et al. 2015). Another common predictor of recurrence is the frequency of previous experiences with the child welfare system, meaning that the more exposure a family has with the system, the more likely they are to be re-reported (Bae et al., 2009; Connell et al., 2009; Janczewski & Mersky, 2016). Other known family predictors for recurrence are high levels of family stress, housing instability, exposure to domestic violence, behavioral health issues, and low socioeconomic status (DePanfilis & Zuravin, 1999; Drake et al., 2003; Drake & Jonson-Reid, 2013; Fluke et al., 2008). Indeed, the majority of families involved in CPS are considered to be in poverty (Font & Maguire-Jack, 2020). Because of the difficulty in measuring poverty, proxy variables are often used such as food stamp receipt (Font & Maguire-Jack, 2020) or Medicaid (Putnam-Hornstein & Needell, 2011). Studies have also found that families with child neglect allegations are more likely to be re-reported or re-substantiated compared to families with other forms of allegations like physical or sexual abuse (Drake et al., 2003; Lipien & Forthofer, 2004). When neglect allegations and poverty are combined, there is increased likelihood of CPS involvement and recurrence (Esposito et al., 2021; Font & Maguire-Jack, 2020). Multiple child risk factors have been linked to higher rates of recurrence, including being younger (especially girls), having developmental or behavior concerns, and having persistently difficult parent-child relationships (Drake et al., 2006; Fluke et al., 2008; Hamilton & Browne, 1999; Marshall & English, 1999). Although the relationship between recurrence and race and/or ethnicity have been thoroughly studied, no consistent findings have emerged (Connell et al., 2007; Maguire-Jack & Font, 2014; Zhang et al. 2013).

1.3. Alternative Response and Recurrence

As noted above, AR was not originally developed to prevent recurrence of child maltreatment, nevertheless it is treated as an intervention to reduce recurrences. In a series of evaluations conducted by the Institute for Applied Research (IAR), the researchers suggested that AR families had lower recurrence rates than TR families. Like the Piper et al., (2019) report, the authors also noted it was challenging to isolate the effect of AR on recurrence due to measurement differences across jurisdictions (Loman & Siegel, 2004, 2013; Siegel et al., 2010). Harries et al. (2015) conducted a retrospective study of AR efficacy in Australia at 10-years post-implementation. The researchers found the number of re-reports and subsequent substantiated investigations decreased and then stabilized at the lower rate during the 10-year study period as a result of the AR policy changes (i.e., reclassification of child maltreatment allegations).

A study conducted by Fluke et al. (2016) examined 6 states that had implemented AR for a minimum of 10 years. The authors found that the 3 states with the highest AR utilization rates yielded significantly lower rates of re-reports than the states with lower AR utilization. Fluke et al. also found lower rates of substantiated re-reports of maltreatment in 5 of the 6 states. On the other hand, a subsequent study of 14 states where states that assigned more than a third (33%) of CPS reports to AR had re-report rates that were equal or higher than traditional investigations (Piper, 2017). Another cross-site report using data from three states again had inconsistent findings. Two states reported lower re-referral rates for AR families, whereas the third state reported no differences in referrals for AR families (Fuller & Zhang, 2017; National Quality Improvement Center on Differential Response in Child Protective Services, 2014). Two additional studies found no differences in re-report between AR and TR families (Conley & Duerr-Berrick, 2010; Fuller & Zhang, 2017), whereas Ji & Sullivan (2016) found that in one jurisdiction AR families had higher rates of service utilization as well as higher rates of subsequent maltreatment reports within one year (Ji & Sullivan, 2016). Overall, the majority of studies examining whether AR reduces recurrence have produced positive findings – AR is as good as CPS and often is better in keeping children safe.

To date, many of the AR studies focus on three primary questions: comparisons of child safety between AR versus TR; differences in service provision, family engagement, and caseworker practice between AR and TR; and cost comparisons of AR versus TR (Fluke et al., 2019; National Quality Improvement Center on Differential Response in Child Protective Services, 2014; Piper et al., 2019). There has been less focus, however, on whether predictors of recurrence (e.g., family size, type of allegation, and socioeconomic status) differ between AR and TR families. This study attempts to add to the literature by looking at the influence of child, family, and jurisdictional factors on recurrence (i.e., re-investigation and substantiated re-investigation) through a multilevel model. The specific questions examined in this study are:

  1. What case and county-level factors predict a re-investigation among AR (RQ1a) and TR (RQ1b) families?

  2. What case and county-level factors predict a substantiated re-investigation among AR (RQ2a) and TR (RQ2b) families?

It is important to note in RQ2 we did not differ on whether the first investigation was substantiated because in this state AR cases cannot result in a substantiated report. Due to the risks that bring families to the attention of CPS and the differentiated risk thresholds for AR and TR families (i.e., lower risk allegations for AR such as neglect; lower risk scores on standardized assessments), we hypothesize the predictors for re-investigations and substantiated re-investigations will be different among the two groups.

2. Method

Data for this study were drawn from one state’s State Automated Child Welfare Information System (SACWIS) across all jurisdictions in the state for all families that received either an AR or TR response from July 1, 2013 – June 30, 2014. Families were followed for 18-months post closure to determine re-investigation and substantiated re-investigation rates. AR was implemented statewide, in five phases, over 15 months. Each county within the state was able to implement AR as it best suited their population, thus most counties had staff that carried both AR and TR cases whereas only four counties had AR specific teams. In general, families were tracked into AR if they were deemed low to moderate risk using the state’s child abuse and neglect needs and risk assessment. Families also could not have sexual abuse or other serious physical (e.g., failure to thrive, death, mental injury) allegations. Other specific nuances to AR policy included any individual with an AR within the last 12 months or a substantiated investigation within 3 years of the last report were automatically tracked into TR. Additionally, AR cases could not exceed 60 days. In order to receive any ongoing services, a family would have to move into the traditional system (i.e., in-home services) but only if a child was “conditionally safe” per the safety assessment or the family was now considered moderate to high risk on the risk assessment.

2.1. Participants

A total of 17,741 families were followed for 18 months post initial AR/TR response. As there is often more than one child involved in an AR or TR, one child from each family was selected using a random number generator. Just under 4% (n=648) of children had missing data on at least one variable. Participants with missing data were removed using listwise deletion. Children with missing data were more likely to have been assigned to AR. Table 1 provides a brief description of the sample characteristics along with bivariate comparisons by response type. Three fifths (60%) of the sample were assigned to AR versus a TR. The sample was evenly split between male and female children. Just over half of the children in the sample (54%) were either Latinx or Black, Indigenous, People of Color (BIPOC). Fourteen percent of the sample was living in a household identified as receiving Medicaid. In their initial responses, just under a third (29%) were investigated for suspected physical abuse, whereas nearly one in ten (9%) were investigated for suspected sexual abuse and 62% were investigated for suspected neglect. In the subsequent re-investigation, one-fifth (20%) were investigated for suspected physical abuse, 15% were investigated for suspected sexual abuse, and almost two-thirds (65%) were investigated for suspected neglect. The mean child age was 8 and on average, there was just under two children in each home (1.8), and the mean number of prior investigations was .3.

Table 1.

Sample Characteristics and Bivariate Comparisons by Response Type (N=17,092).

Total % Traditional Response Alternate Response
% %
Response Traditional 40.2
Alternative 59.8
Gender Female 49.6 54.8 47.4
Male 50.4 45.2 52.6
BIPOC or Latinx Yes 54.3 51.2 56.4
No 45.7 48.8 43.6
Medicaid Yes 14.1 24.6 7.0
No 85.9 75.4 93.0
Initial TR/AR Allegation Physical Abuse 29.1 21.0 34.5
Sexual Abuse 9.0 22.3 0.0
Neglect 62.0 56.7 65.5
Substantiated Re-investigation Allegation Physical Abuse 19.6 19.9 19.3
Sexual Abuse 15.5 16.3 14.2
Neglect 64.9 63.8 66.6
Risk Level at Initial TR/AR Medium/High 29.5 52.9 13.9
Low/No 70.5 47.1 86.1
Mean Mean Mean
(SD) (SD) (SD)
Number of Children 1.84 1.91 1.79
(1.30) (1.37) (1.24)
Child Age 7.97 7.73 8.12
(4.94) (5.20) (4.76)
Prior Investigations .30 .47 .18
(.74) (.95) (.52)
Open in a new tab

2.2. Measures

2.2.1. Reported re-investigation (outcome variable; RQ1).

Reported re-investigation is a dichotomous variable indicating whether or not a family received a subsequent investigation (0 = No, 1= Yes).

2.2.2. Substantiated re-investigation (outcome variable; RQ2).

Substantiated re-investigation is a dichotomous variable indicating whether the subsequent investigation was substantiated (0 = No, 1= Yes).

2.2.3. Child and family characteristics.

The majority of the child-level characteristics were dichotomous and were measured during the initial AR or TR. The variables examined included gender (female = 0, male = 1), child age at the time of the report (continuous), number of children in the home (continuous), and child guardian race and ethnicity (1 = BIPOC or Latinx, 0 = Non-Latinx White). We made the choice to combine race and ethnicity into a single variable in order to address missing data. Nearly all families who identified as Latinx had missing data for race and vice versa. Combining the two variables allowed us to maintain a larger sample while still including the maximum amount of information regarding race and ethnicity. If there was missing data for the child the data from parent race/ethnicity was used. In addition, we examined variables indicating whether a family was referred to AR (=1 vs TR = 0), the number of prior investigations associated with the child (continuous), and whether the child received health insurance through Medicaid (0 = No, 1= Yes). A series of dummy variables were created to indicate whether the initial and/or the re-investigation allegation report was for physical abuse, sexual abuse, or neglect (reference group=neglect). It is important to note that families with an initial allegation of sexual abuse were automatically referred to TR due its high-risk level, thus sexual abuse was not included as a model parameter for RQ1a. The risk group variable was based on the scores of the states’ risk assessment. The final scores are placed into ‘high’, ‘moderate’, ‘low’, or ‘no’. We created a dichotomous variable to indicate whether a child was deemed to be at high or medium risk (=1 vs. low- or no-risk = 0). This state’s risk assessment specifically provides guidance for caseworkers on what families need services and/or what services are needed to reduce overall risk. Caseworkers are expected to complete the tool at the close of the investigation, or AR, per state statute. All child-level variables were group-mean centered.

2.2.4. Jurisdiction-level characteristics.

Following the methodology suggested by Raudenbush & Bryk (2002) jurisdiction-level variables were defined by calculating the jurisdiction-wide percentages or mean values of all level-1 model parameters (e.g., percentage of children who were assigned to AR per jurisdiction, mean number of prior reports per jurisdiction). Children were allocated the values for (i.e., nested in) the jurisdiction where they were housed during the relevant re-investigation or substantiated re-investigation. All jurisdiction-level variables were grand-mean centered.

2.3. Data Analyses

All analyses for the present study were conducted using SAS, Version 9.3. The present study examined the predictors of a reported re-investigation and/or a substantiated re-investigation among families assigned to AR and TR across all jurisdictions in one mid-Atlantic state. We conducted a series of two-level multilevel generalized linear mixed models with the GLIMMIX module for SAS 9.3 using a logit link to accommodate binary outcomes (Dai et al., 2006; METHOD=LAPLACE). Between-jurisdiction differences in the outcome variable were evaluated through a varying (random) intercept model. All child- and jurisdiction-level parameters were treated as constant (fixed) effects (Gelman & Hill, 2006). All assumptions of the multilevel logistic regression were assessed. We included grand-mean centered jurisdiction-level counterparts of all child-level factors, which were group mean centered, as a means of disentangling group (jurisdiction-level) effects from individual (child-level) effects, (Bafumi & Gelman, 2007; Raudenbush & Bryk, 2002). We group ran Pearson correlations to assess collinearity between study variables. After centering, the highest correlation was r =.295 suggesting that there was no problematic collinearity among the study variables.

3. Results

3.1. Reported Re-investigations in Alternative (RQ1a) and Traditional Response (RQ1b)

The z-test for the covariance parameters for RQ1a (z = 2.90, p =.002) and RQ1b (z = 1.97, p =.024) both indicated statistically significant between-jurisdiction variation in reported re-investigations, providing justification for the use of MLM techniques (Hox, 2002). Results for the multilevel model fitted to evaluate the contributions of child and jurisdiction-level factors on child welfare report re-investigations are presented in Table 2.

Table 2.

Summary of Results for the Multilevel logit Models Fitted to Evaluate the Contributions of Individual and County-level Factors on Re-investigations and Substantiated Re-investigation Among Families Assigned to Alternative and Traditional Response (N=17,092)

Reported Re-investigations a. Substantiated Re-investigation b.
Alternative Response (RQ1a) Traditional Response (RQ1b) Alternative Response (RQ2a) Traditional Response (RQ2b)
(n= 10,300) (n= 6,792) (n= 1,023) (n= 1,049)
OR p OR p OR p OR p
Level-1 Fixed Effects
 Intercept <.0001 <.0001 .286 .019
 Child Gender (Female) .886 .081 .925 .277 .968 .807 1.249 .095
 Child Age .958 <.0001 .987 .070 .968 .031 .989 .435
 BIPOC or Latinx 1.135 .113 1.057 .487 .983 .914 .914 .559
 Number of Children 1.129 <.0001 1.072 .005 1.147 .006 1.037 .401
 Prior Reports 1.444 <.0001 1.295 <.0001 .982 .807 .983 .750
 Medicaid 4.445 <.0001 2.195 <.0001 1.580 .003 1.568 .001
 Physical Abuse c. .953 .578 1.020 .832 .863 .411 .522 <.0001
 Sexual Abuse c. † .734 .008 1.107 .643 1.471 .053
 Medium/High Risk 1.610 <.0001 1.321 <.0001 1.320 .073 .681 .009
Level-2 Fixed Effects
 % Female 1.194 .527 .733 .448 1.269 .575
 Mean Child Age 1.561 .011 1.040 .843 .520 .029
 % BIPOC or Latinx 1.087 .048 1.014 .820 .865 .037
 Mean # of Children 1.054 .241 1.090 .820 1.087 .256
 Mean Prior Reports 1.208 <.0001 .989 .205 .961 .630
 % Medicaid 1.103 .299 .985 .911 1.085 .612
 % Physical Abuse .998 .986 1.250 .170 1.034 .828
 % Sexual Abuse † .433 .149 .740 .050
 % Medium/High Risk 1.310 <.0001 1.013 .896 .746 .013
Covariance Parameter Estimates
 Intercept .020 d. .076 .040 d. .102 .001 d. .488 .114 d. .213 e.
Open in a new tab

Note.

a.

The reference category is 1.00: Reported re-investigation.

b.

The reference category is 1.00: Substantiated re-investigation.

c.

RQ1a and b = allegation from index event, RQ2a and b = allegation from recurrence event.

d.

Covariance parameter estimate coefficient.

e.

There was insufficient between-jurisdiction variance in the null model to support the inclusion of level-2 parameters in the final model.

†

Families referred to AR could not have an index case of reported for suspected sexual abuse.

3.1.1. Child and family level factors.

Families, in AR, but not TR (pTR =.070), who had a child younger than the mean child age for the respective group (ORAR=.958, p<.0001) were less likely to have a reported re-investigation. In addition, families in TR, whose initial report was for sexual abuse had lower odds of a re-investigation compared to families with an initial report of neglect (ORTR=.734, p =.008). Regardless of response type, families with above average number of children in their homes (ORAR=1.129, p<.0001; ORTR=1.072, p =.005) and families with an above average number of prior investigations (ORAR=1.444, p<.0001; ORTR=1.295, p<.0001) were more likely to have a reported re-investigation. Families receiving health coverage through Medicaid (ORAR=4.445, p<.0001; ORTR=2.195, p<.0001), and designated as medium or high risk (vs. low or no risk; ORAR=1.610, p<.0001; ORTR=1.321, p<.0001) were more likely to have a reported re-investigation. There was no relationship between gender (pAR=.081; pTR=.277), being BIPOC/Latinx (pAR=.113; pTR=.487) or having an index report of physical abuse (vs. neglect; pAR=.578; pTR=.832) and the odds of a reported re-investigation.

3.1.2. Jurisdiction-level factors.

Children who were assigned to AR and lived in jurisdictions with a mean child age above the state average were more likely to have a reported re-investigation (ORAR=1.561, p<.01). Children who were assigned to AR and lived in jurisdictions with an above average proportion of BIPOC/Latinx children on their caseload (ORAR=1.087, p =.04), an above average rate of mean prior reports (ORAR=1.208, p<.0001), or an above average proportion of the population designated as medium or high risk (vs. low or no risk; ORAR=1.310, p<.0001) were more likely to have a reported re-investigation but not on the other factors: number of children, Medicaid receipt, or type of allegation. For families assigned to TR, there was no relationship between jurisdiction level measures and the odds of re-investigation for child gender (pTR=.277), child age (pTR=.843), BIPOC/Latinx (pTR=.820), number of children per household (pTR=.956), prior reports (pTR=.989), allegation type [physical (pTR=.170) or sexual (pTR=.149) abuse], or level of risk (pTR=.896). For families assigned to TR, there was no relationship between jurisdiction level measures and the odds of re-investigation for child gender (pTR=.277), child age (pTR=.843), BIPOC or Latinx (pTR=.820), number of children per household (pTR=.956), prior reports (pTR=.989), allegation type [physical (pTR=.170) or sexual (pTR=.149) abuse], or level of risk (pTR=.896).

3.2. Substantiated Re-investigations in AR (RQ2a) and TR (RQ2b)

The z-test for the covariance parameters for RQ2a indicated statistically significant between-jurisdiction variation in substantiated re-investigations (z = 2.09, p = .037), providing justification for the use of MLM techniques (Hox, 2002). The z-test for the covariance parameters for RQ2b was not statistically significant (z = 1.17, p = .243). As there was no significant between-jurisdiction variation in the dependent variable we did not include any level-2 predictors for RQ2b. We did however, out of an abundance of caution, maintain the MLM analysis structure in order to address the nested structure of the data and prevent overestimation of the model’s standard errors (Hox, 2002). Results for the multilevel model fitted to evaluate the contributions of child and jurisdiction-level factors on child welfare report recurrences are presented in Table 2 (see Table 2 above).

3.2.1. Child and family-level factors.

Families assigned to AR and had children older than average (ORAR=.968, p=.031) were less likely to have a substantiated re-investigation; however, those families who had an above average number of children in their homes (ORAR=1.147, p=.006) were more likely to have a substantiated re-investigation. There was no relationship between suspected physical (pAR=.411), suspected sexual abuse (pAR=.643), or risk level (pAR=.073) and the odds of a substantiated re-investigation for AR families. Among AR and TR families receiving health coverage through Medicaid both were more likely to have a substantiated re-investigation (ORAR=1.580, p= .003; ORTR=1.568, p =.001) but there was no relationship between gender (pAR=.807; pTR=.095), being BIPOC or Latinx (pAR=.914; pTR=.559), or number of prior investigations (pAR=.807; pTR=.750). Families assigned to TR and were referred for suspected physical abuse (vs. neglect; ORTR=.522, p<.0001) or were assessed to be medium or high risk (vs. low/medium risk; ORTR=.681, p < .01) had decreased odds of a substantiated re-investigation. There was no relationship between the child’s age and the odds of a substantiated re-investigation (pTR=.401) for TR families.

3.2.2. Jurisdiction-level Factors.

Families initially assigned to AR who lived in jurisdictions with a mean child age above the state average (ORAR=.520, p=.029), lived in jurisdictions with above average concentrations of BIPOC and Latinx cases (ORAR=.839, p=.037), and those families living in jurisdictions with an above average rate of families being assessed as medium or high risk (ORAR=.746, p=.013) were less likely to have a substantiated re-investigation. There was no relationship between the jurisdiction-level proportion of female children (pAR=.575), the mean number of children per household (pAR=.256) and the odds of a substantiated re-investigation. Jurisdiction-level mean prior investigations (pAR=.630), the percentage of children receiving health coverage through Medicaid (pAR=.612), and the proportion of children who had been referred for physical abuse (pAR=.828) or sexual abuse (pAR=.050) were not significant for a substantiated re-investigation. There was insufficient between-jurisdiction variance in the dependent variable in RQ2b to justify including level-2 parameters in the final model.

4. Discussion

The purpose of this study was to describe what factors predicted an initial re-investigation and a substantiated re-investigation after first receiving either an AR or TR in CPS. Before discussing the findings, it is important to advise against directly comparing AR and TR as there are some notable differences between the groups, such as families who received an AR could not have severe physical abuse or sexual abuse allegations for their initial response; this could be an allegation for a substantiated re-investigation, however.

4.1. Research Question 1 - Predictors of a Reported Re-investigation

Despite our hypothesis that there would be a difference between the predictors of re-investigation for AR and TR families, we identified few differences. The only significant differences for AR were age of the child and for TR families the type of allegation (sexual abuse). For both AR and TR families, the number of children in the family, number of prior investigations, Medicaid receipt, and risk levels were significant predictors of a reported re-investigation. At the county level age, race/ethnicity, prior reports, and risk levels were significant predictors of re-investigation for AR families; there were no significant predictors at the jurisdictional level for TR families.

The finding from the current study that TR families with more children were more likely to have a re-investigation is supported by previous literature (Bae et al., 2009; Fuller & Nieto, 2009; Drake et al., 2006). The present findings support the notion that having more children in the household may correlate to more stress, which may in turn lead to increased maltreatment rates (Fluke et al., 2008). Our finding that previous exposure to the system (i.e., number of previous investigations) impacts recurrence rates also aligns with findings from previous studies (Fluke et al., 2008; Hélie et al., 2013). Although we are unfamiliar with studies that have looked at previous CPS involvement as a predictor of re-investigation for AR families, we believe that these families may be impacted by this type of surveillance, particularly where previous reports, regardless of CPS track (i.e., AR or TR) are kept for a number of years. Ideally, the needs of these families are met during the AR through concrete support and services. However, some family situations are particularly complex and the 60-day timeframe and a lack of ongoing options for AR families in this state may have prevented the initial situation from being adequately addressed. Finally, allegation type has not presented with consistent findings (Jonson-Reid et al., 2010; Kahn & Schwalbe, 2010) so our finding about sexual abuse lowering the odds for a re-investigation in TR families was not expected. Nonetheless, an explanation for this finding could rest in the overall difficulties of obtaining enough evidence to indicate a sexual abuse allegation (Kellogg, 2005).

Our findings that TR families deemed higher risk as indicated by a state’s risk assessment are more likely to experience a reported re-investigation is aligned with previous research findings (Brown et al., 2020). It is important to note that in practice the risk assessments are often used for more than just determining future maltreatment and are sometimes used to determine immediate safety of a child (Hughes & Rycus, 2006; Knoke & Trocme, 2005). Previous research assessing the accuracy of risk assessments suggest moderate predictive utility, however, our findings suggest that they may be a useful tool for identifying TR families at risk for re-investigation or re-report (Van der Put et al., 2017). This same suggestion does not extend to AR families. The proportion of families who received medium-high risk ratings was close to 14% suggesting a need to reexamine the implementation of the policy and the policy’s relationship to the risk tools (i.e., what families are most appropriate for the AR track). These particular AR families (with medium-high risk) might have additional stressors that were beyond the scope of AR practice and thus, required more services and resources than the AR workers could provide. AR caseworkers have been shown to focus only on the contents of the current report and not expand their scope to include additional services (First Author, 2020).

The finding of an increased risk for re-investigation among families who received Medicaid aligns with previous research that used receipt of public assistance as an indicator of financial need (Connell et al., 2007; Lee et al., 2012). Although the type of assistance included in the analyses (e.g., Medicaid, Supplemental Nutrition Assistance Program (SNAP) Temporary Aid for Needy Families) may inflate the results. Our results, however, should be interpreted with caution as Medicaid is at best a rough estimate of poverty, and in this state knowing whether the family receives Medicaid falls on the caseworker asking and subsequently adding it to the family’s case record. Still, Medicaid receipt produced the highest odds of a reported re-investigation among the model parameters for both AR (RQ1a) and TR (RQ1b). Given that healthcare policy (i.e., Patient Protection and Affordable Care Act) and other social policies have shifted to include higher income families for services it may prove beneficial to explore the relationship between families with more economic resources and determine if there are differences with their involvement in child welfare services.

At the county level, only AR families had significant findings including jurisdiction-level measures of age, race/ethnicity, number of investigations, and families rated as medium-high risk. This is best interpreted as AR families who lived in counties with above average age, higher concentration of BIPOC/Latinx cases, higher than average rates of prior investigations or lived in counties with a higher than average concentration of families rated as medium to high risk had higher odds of having a re-investigation. These findings could be related to county level differences in the implementation of AR in this state. Future research should conduct between-county comparisons controlling for implementation to determine if the present findings are sustained at the individual county level, particularly with race/ethnicity as these findings continue to be widely debated at the systemic level (Barth et al., 2020; Detlaff et al., 2020).

4.2. Research Question 2: Predictors of a Substantiated Re-investigation

Unlike the initial response, the factors that predicted a substantiated re-investigation were different for AR and TR families. For families who initially received an AR, only two predictors were significant: child age lowered the odds of a substantiated re-investigation whereas having more children in the home increased the odds of a substantiated re-investigation. For families who initially received TR having a physical abuse allegation and medium – high risk levels lowered a family’s odds of having a substantiated re-investigation. For both AR and TR families, receipt of Medicaid increased their odds of receiving a substantiated re-investigation. The county-level variables were different in that the findings flipped from the first research question and the odds of substantiated re-investigation were lower for AR families who lived in counties with children above the state average age, had above average concentrations of BIPOC/Latinx families, and had above average rates of medium or high risk.

The findings for older children and more children in the home, as well as Medicaid receipt are all in line with current research, although future research could continue teasing out if these differences are sustained between track assignments. Our findings, however, regarding physical abuse for TR families are worth discussing further. In general, the type of allegation has consistently related to substantiation findings (Connell et al., 2009; Fluke et al., 2008; Jonson-Reid et al., 2010; Kahn & Schwalbe, 2010). What is not clear is whether an additional report of physical abuse yields a substantiation. Although Jonson-Reid et al. (2010) found that physical abuse had lower recurrence rates for subsequent reports, physical abuse, in general, was not more likely to be substantiated in future reports. Another explanation, which is relevant to this study as we did not control for previous allegation, is that families do not always return to the system with the same allegations (Jonson-Reid et al., 2003). Also unclear is why families with medium-high risk levels had lower odds of receiving a substantiated recurrence. In general, higher risk levels result in further system action even if that is another report or investigation (Russell et al., 2018). It is worth exploring these differences at the individual county level to determine if the result is replicated.

With regards to the county level variables, age, race/ethnicity, and risk was significant for families who initially received an AR but in the opposite direction from our first research question. The odds of a substantiated re-investigation were lower for previous AR families who lived in counties with older children, higher concentration of BIPOC/Latinx families, and higher than average concentration of families rated as medium to high risk. This could mean that there are some interacting unmeasured factors at the jurisdictional level. It is possible that there is a confluence of family constellation and re-investigation that is confounding the results or that there are differences in practice at the jurisdictional level that are impacting substantiated re-investigations. These findings underscore the need to further investigate cross-jurisdictional differences of AR within individual states.

5. Implications

Policy and Practice.

Nationally, AR has been implemented in over 30 jurisdictions (Piper et al., 2019). Most previous research has shown decreased or similar levels of recurrence for families assigned to AR compared to TR (Conley & Duerr-Berrick, 2010; Loman & Siegel, 2015). Given our results suggest there are few differences in what predicts a family returning to the system, regardless of response, it is worth considering if the number of resources (e.g., changing the policy process, training caseworkers, explaining the track differences to families and community members) dedicated to AR is worth it? This is not meant to undermine the studies that suggest recurrences are reduced because that is relatively consistent with AR. It may be prudent, however, to further examine contextual differences among jurisdictions such as when caseworkers must conduct both AR and TR. Caseworkers in this state reported having difficulty switching between response approaches (i.e., focusing on developing family rapport in AR versus collective forensic evidence in TR; First Author, 2020) and could benefit from increased knowledge of how to best handle these situations so they can appropriately address family needs. It may also be worth investing in a cost effectiveness analysis, which are essentially “return on investment” processes for the public sector (Crowley et al., 2018). Public sector decisions are complex, and these types of analyses help determine society’s “profit” from investing in AR models. Finally, it may be beneficial for child welfare researchers and administrators to consider operationalizing success using metrics other than recurrence rates. Moreover, it would be particularly useful for jurisdictions to refrain from comparing their outcomes to those states in which they have little in common (Drake, 2013). One reason AR is attractive is its flexible framework, but these continued comparisons disallow an understanding of what “works best” for that individual system.

Research.

Recurrence is a useful outcome to measure when determining whether a child welfare agency is optimally performing; however, the findings from this study and others demonstrate that there are many factors that lead to recurrence, and type of response, at least at the child and family level, did not indicate much difference. It is prudent to study an entire organization to fully comprehend the strengths and weaknesses. Obtaining and operationalizing organizational factors may be challenging but it may be the piece that is most needed to accurately assess the successes of AR. Another area of research that has received less attention is the screening and intake processes used by child welfare systems. There has been some research in this area (Ji & Sullivan, 2016), however, this study did not examine which factors predicted the type of approach (AR vs. TR) a family will receive. The burgeoning field of AR utilization rates (Fluke et al., 2019; Piper, 2017) and recurrence are also helping to shed light on determining which families are best suited for a particular track, beyond risk level assessment and type of allegation. Finally, given the differences in our jurisdictional (second) level findings (i.e., the predictors from RQ1 flipped with RQ2) future research would benefit from doing additional multilevel modeling analyses where we can control for cross-level interactions.

There must also be continued consideration about all family and case factors given the historical trend of systemic racism in CPS. Although race and ethnicity were predictive only in RQ1 at the county level, other studies have found Black and Latinx families with neglect allegations (i.e., lower risk) are less likely to receive an AR (Choi et al., 2021). Another study found Black families living in areas with high substantiation rates were also less likely to receive an AR (Connell, 2020). These findings are antithetical to the principles of AR that suggest low to moderate risk are the target population; the race and ethnicity of a family should not factor into the track assignment. These findings, in addition to other studies focusing on child welfare experiences among families of color (Merritt, 2021; Roberts, 2009) underscore the need to continue conducting research that uses multiple methods (e.g., quantitative and qualitative) with multiple voices (e.g., CPS families, CPS administrators and caseworkers) but makes race and ethnicity the primary variables of interest.

6. Strengths and Limitations

As with all studies, there are limitations worth highlighting from this study. First, was our definition of recurrence. Although re-investigation is not unusual to use to determine the rate in which families come back into the system, there is an assumption that an unsubstantiated investigation equals a lack of maltreatment (Drake, 1996). This is often not the case but rather the investigator was unable to find sufficient evidence to justify further intervention from CPS. Indeed, this gray area was the rationale for most studies moving to re-report. In our second research question we used substantiated re-investigations. The purpose of using substantiation was to indicate a definitive end point to a case and to determine if there were any differences between a family’s initial track assignment and future re-investigations. However, the use of substantiation is concerning as it has multiple administrative purposes, including determining services needs and court involvement related to removal. This determination also relies on the judgment of a caseworker or team of caseworkers based on a series of tools that vary widely from state to state (Drake & Jonson-Reid, 2000). Another concern with using substantiated re-investigations is AR families in this state do not receive an initial substantiation. Although we were aware if a family’s initial report was AR, we could not determine if this influenced a subsequent substantiation as the first report did not have a final determination. Second, because families were not randomly assigned to treatment conditions (i.e., AR vs. TR) it is not possible to make any causal conclusions; this reinforces the need to continue using study designs with a randomized control approach. Our final limitation was our reliance on administrative data. The use of these data has grown exponentially to determine trends and outcomes of agency work but they are also wrought with challenges such as missing data, predetermined variables, and bias (Capatosto, 2017; English et al., 2000). This study was no exception to these challenges although the information provided does offer additional insights to the ongoing issue of recurrence in CPS, specifically within AR.

AR continues to be an area of current child welfare policy and practice, thus any research that adds to this subject is necessary as more jurisdictions choose to implement AR. Moreover, recurrence is an ongoing and expensive issue for all child welfare jurisdictions. A strength of this study was isolating AR and digging deeper at the recurrence dilemma. The standard time families are followed is six-months post-investigation. This is because most families return within this time frame (Fluke et al., 2008) but also because using national datasets can limit the follow-up period. By using our state data system, we were able to follow families for 18-months, which is typically the time in which recurrence rates will level out. Another strength of this study was the use of data in one state as opposed to individual jurisdictions as well as the analytical approach. By using a multilevel modeling approach, we were able to account for the effect of nesting within jurisdictions on the estimation of standard errors. Finally, by group-mean centering the individual-level parameters and including their grand-mean centered jurisdiction-level counterparts we were able to isolate the effects of both the individual- and jurisdiction-level characteristics on the outcomes of interest.

7. Conclusion

Although this study is not the first to focus on differences in CPS approaches, it is the first to use multilevel modeling to control for group differences at the county level and compare predictors of re-investigations and substantiated re-investigations in both AR and TR cases, as well as follow families for an extended period of time (e.g., 18-months). We understand the lack of a control group does not support causal inferences and that we were limited in our predictive variables due to using administrative data. However, we believe our results provide new information about how AR and TR families, at least in this state, might be more similar than they are different. This further emphasizes the need to continue digging into the specifics of what can make an optimal AR track.

Highlights.

  • Alternative response is a widely used practice in child protective services

  • Despite use of AR, families continue to return to the child protective system

  • Family return, or recurrences, are not substantially different for individual families

  • County level differences of AR require further investigation

Footnotes

Credit Author Statement

Stacey L. Shipe: conceptualization, methodology, investigation, data curation, writing – original draft, writing – review and editing, visualization, supervision, funding acquisition

Mathew C. Uretsky: methodology, software, validation, formal analysis, investigation, writing – review and editing

Terry V. Shaw: validation, resources, writing – review and editing

Conflicts of Interest Statement

The authors whose names are listed immediately below certify that they have NO affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or non-financial interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed in this manuscript.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  1. Bae HO, Solomon PL, Gelles RJ, & White T. (2009). Effect of child protective services system factors on child maltreatment rereporting. Child Welfare, 89(3), 33–55. [PubMed] [Google Scholar]
  2. Barth RP, Jonson-Reid M, Greeson JK, Drake B, Berrick JD, Garcia AR, ... & Gyourko JR (2020). Outcomes following child welfare services: what are they and do they differ for black children?. Journal of Public Child Welfare, 14(5), 477–499. [Google Scholar]
  3. Benbenishty R, Davidson-Arad B, López M, Devaney J, Spratt T, Koopmans C, ... & Hayes D. (2015). Decision making in child protection: An international comparative study on maltreatment substantiation, risk assessment and interventions recommendations, and the role of professionals’ child welfare attitudes. Child Abuse & Neglect, 49, 63–75. [DOI] [PubMed] [Google Scholar]
  4. Bickel R. (2007). Multilevel analysis for applied research: It’s just regression! New York, NY: The Guilford Press. [Google Scholar]
  5. Byrne BM (2013). Structural equation modeling with Mplus: Basic concepts, applications, and programming. New York, NY: Routledge Publishers. [Google Scholar]
  6. Cameron G, & Freymond N. (2015). Accessible service delivery of child welfare services and differential response models. Child abuse & neglect, 39, 32–40. [DOI] [PubMed] [Google Scholar]
  7. Capatosto K. (2017). Foretelling the future: A critical perspective on the use of predictive analytics in child welfare. Columbus, OH: Kirwan Institute for the Study of Race and Ethnicity. [Google Scholar]
  8. Carnochan S, Rizik-Baer D, & Austin MJ (2013). Preventing re-entry to foster care. Journal of Evidence-Based Social Work, 10(3), 196–209. [DOI] [PubMed] [Google Scholar]
  9. Carnochan S, Rizik-Baer D, & Austin MJ (2013). Preventing the recurrence of maltreatment. Journal of Evidence-based Social Work, 10(3), 161–178. [DOI] [PubMed] [Google Scholar]
  10. Casanueva C, Tueller S, Dolan M, Testa M, Smith K, & Day O. (2015). Examining predictors of re-reports and recurrence of child maltreatment using two national data sources. Children and Youth Services Review, 48, 1–13. [Google Scholar]
  11. Chaffin M, Bard D, Hecht D, & Silovsky J. (2011). Change trajectories during home-based services with chronic child welfare cases. Child Maltreatment, 16(2), 114–125. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Child Welfare Information Gateway (2017). Child Maltreatment 2015: Summary of Key Findings. Washington, DC: U.S. Department of Health and Human Services. [Google Scholar]
  13. Cross TP, & Casanueva C. (2009). Caseworker judgments and substantiation. Child Maltreatment, 14(1), 38–52. [DOI] [PubMed] [Google Scholar]
  14. Crowley DM, Dodge KA, Barnett WS, Corso P, Duffy S, Graham P, ... & Karoly LA (2018). Standards of evidence for conducting and reporting economic evaluations in prevention science. Prevention Science, 19(3), 366–390. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Conley A. (2007). Differential response: A critical examination of a secondary prevention model. Children & Youth Services Review. 2007:29:1454–1468. [Google Scholar]
  16. Conley A, & Duerr Berrick J. (2010). Community-based child abuse prevention: Outcomes associated with a differential response program in California. Child maltreatment, 15(4), 282–292. [DOI] [PubMed] [Google Scholar]
  17. Connell CM, Vanderploeg JJ, Katz KH, Caron C, Saunders L, & Tebes JK (2009). Maltreatment following reunification: predictors of subsequent Child Protective Services contact after children return home. Child Abuse & Neglect: The International Journal, 33(4), 218–228. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Coohey C, Johnson K, Renner LM, & Easton SD (2013). Actuarial risk assessment in child protective services: Construction methodology and performance criteria. Children and Youth Services Review, 35(1), 151–161. [Google Scholar]
  19. Cuccaro-Alamin S, Foust R, Vaithianathan R, & Putnam-Hornstein E. (2017). Risk assessment and decision making in child protective services: Predictive risk modeling in context. Children and Youth Services Review, 79, 291–298. [Google Scholar]
  20. DePanfilis D, & Zuravin SJ (2001). Assessing risk to determine the need for services. Children and Youth Services Review, 23(1), 3–20. [Google Scholar]
  21. DePanfilis D, & Zuravin SJ (1999). Predicting child maltreatment recurrences during treatment. Child Abuse & Neglect, 23(8), 729–743. [DOI] [PubMed] [Google Scholar]
  22. Dettlaff AJ, & Boyd R. (2020). Racial disproportionality and disparities in the child welfare system: Why do they exist, and what can be done to address them?. The ANNALS of the American Academy of Political and Social Science, 692(1), 253–274. [Google Scholar]
  23. Drake B. (1996). Unraveling “unsubstantiated”. Child Maltreatment, 1(3), 261–271. [Google Scholar]
  24. Drake B, Jolley JM, Lanier P, Fluke J, Barth RP, & Jonson-Reid M. (2011). Racial bias in child protection? A comparison of competing explanations using national data. Pediatrics, 127(3), 471–478. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Drake B., Jonson-Reid M. (2013). Poverty and child maltreatment. In Korbin J, Krugman RD (Eds.), Handbook of child maltreatment (Vol. 2, pp. 131–148). New York, NY: Springer. [Google Scholar]
  26. Drake B, & Jonson-Reid M. (2014). Poverty and child maltreatment. In Korbin J, & Krugman RD (Eds.), Handbook of child maltreatment (Vol. 2). Netherlands: Springer [Google Scholar]
  27. Drake B, Jonson-Reid M, Way I, & Chung S. (2003). Substantiation and recidivism. Child Maltreatment, 8(4), 248–260. [DOI] [PubMed] [Google Scholar]
  28. English DJ, Wingard T, Marshall D, Orme M, & Orme A. (2000). Alternative responses to child protective services: Emerging issues and concerns. Child Abuse & Neglect, 24(3), 375–388. [DOI] [PubMed] [Google Scholar]
  29. English DJ, Marshall DB, Brummel S, & Orme M. (1999). Characteristics of repeated referrals to child protective services in Washington State. Child Maltreatment, 4(4), 297–307. [Google Scholar]
  30. Farrow F, Notkin S, Derezotes D, & Miller O. (2011, December). Race Equity in Child Welfare: Key Themes, Findings and Perspectives. In Research Synthesis on Child Welfare Disproportionality and Disparities. Baltimore, MD: Annie E. Casey Foundation. [Google Scholar]
  31. Fluke JD (2008). Child protective services rereporting and recurrence—Context and considerations regarding research. Child Abuse & Neglect, 32(8), 749–751. [DOI] [PubMed] [Google Scholar]
  32. Fluke JD, Harlaar N, Brown B, Heisler K, Merkel-Holguin L, & Darnell A. (2019). Differential Response and Children Re-Reported to Child Protective Services: County Data From the National Child Abuse and Neglect Data System (NCANDS). Child Maltreatment, 24(2), 127–136. [DOI] [PubMed] [Google Scholar]
  33. Fluke JD, & Hollinshead D. (2003). Child maltreatment recurrence: A leadership initiative of National Resource Center on Child Maltreatment. Duluth, GS: National Resource Center on Child Maltreatment. [Google Scholar]
  34. Fluke JD, Jones-Harden B, Jenkins M, & Ruehrdanz A. (2010). Research synthesis on child welfare disproportionality and disparities. Washington, DC: American Humane Association and Center for the Study of Social Policy. [Google Scholar]
  35. Fluke JD, Shusterman GR, Hollinshead DM, & Yuan YYT (2008). Longitudinal analysis of repeated child abuse reporting and victimization: Multistate analysis of associated factors. Child Maltreatment, 13(1), 76–88. [DOI] [PubMed] [Google Scholar]
  36. Font SA, & Maguire-Jack K. (2020). It’s not “Just poverty”: Educational, social, and economic functioning among young adults exposed to childhood neglect, abuse, and poverty. Child Abuse & Neglect, 101, 104356. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Fuller T, & Nieto M. (2009). Substantiation and maltreatment rereporting: A propensity score analysis. Child Maltreatment, 14(1), 27–37. [DOI] [PubMed] [Google Scholar]
  38. Gelman A. & Hill J. (2007). Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge, UK: Cambridge University Press. [Google Scholar]
  39. Hamilton CE, & Browne KD (1999). Recurrent maltreatment during childhood: A survey of referrals to police child protection units in England. Child Maltreatment, 4(4), 275–286. [Google Scholar]
  40. Harries M, Cant RL, Bilson A, & Thorpe D. (2015). Responding to information about children in adversity: Ten years of a differential response model in Western Australia. Child Abuse & Neglect, 39, 61–72. [DOI] [PubMed] [Google Scholar]
  41. Hélie S, Laurier C, Pineau-Villeneuve C, & Royer MN (2013). A developmental approach to the risk of a first recurrence in child protective services. Child Abuse & Neglect, 37(12), 1132–1141. [DOI] [PubMed] [Google Scholar]
  42. Hindley N, Ramchandani PG, & Jones DP (2006). Risk factors for recurrence of maltreatment: a systematic review. Archives of disease in childhood, 91(9), 744–752. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Hollinshead DM (2012). Do outcomes reflect performance, demographic dividends, or policy artifacts? A study of the effects of child welfare policies, practices, and demographics on state child maltreatment recurrence rates. (Doctoral dissertation) Available from ProQuest Dissertation and Theses database (UMI No. 3539943) [Google Scholar]
  44. Hughes RC, Rycus JS, Saunders-Adams SM, Hughes LK, & Hughes KN (2013). Issues in differential response. Research on Social Work Practice, 23(5), 493–520. [Google Scholar]
  45. Hughes RC, & Rycusa JS (2006). Issues in risk assessment in child protective services. Journal of Public Child Welfare, 1(1), 85–116. [Google Scholar]
  46. Janczewski CE, & Mersky JP (2016). What’s so different about differential response? A multilevel and longitudinal analysis of child neglect investigations. Children and youth services review, 67, 123–132. [Google Scholar]
  47. Ji D, & Sullivan R. (2016). The Manifest and Latent Functions of Differential Response in Child Welfare. Research on Social Work Practice, 26(6), 641–652. [Google Scholar]
  48. Jonson-Reid M, Emery CR, Drake B, & Stahlschmidt MJ (2010). Understanding chronically reported families. Child Maltreatment, 15(4), 271–281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Jonson-Reid M, Drake B, Kohl P, Guo S, Brown D, McBride T, ... & Lewis E. (2017). What do we really know about usual care child protective services?. Children and Youth Services Review, 82, 222–229. [Google Scholar]
  50. Kahn JM, & Schwalbe C. (2010). The timing to and risk factors associated with child welfare system recidivism at two decision-making points. Children and Youth Services Review, 32(7), 1035–1044. [Google Scholar]
  51. Kaplan C, & Merkel-Holguin L. (2008). Another look at the national study on differential response in child welfare. Protecting Children, 23(1&2), 5–21. [Google Scholar]
  52. Kellogg N. (2005). The evaluation of sexual abuse in children. Pediatrics, 116(2), 506–512. [DOI] [PubMed] [Google Scholar]
  53. Knoke D, & Trocme N. (2005). Reviewing the evidence on assessing risk for child abuse and neglect. Brief Treatment and Crisis Intervention, 5(3), 310. [Google Scholar]
  54. Lee S, Jonson-Reid M, & Drake B. (2012). Foster care re-entry: Exploring the role of foster care characteristics, in-home child welfare services and cross-sector services. Children and Youth Services Review, 34(9), 1825–1833. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Lipien L, & Forthofer MS (2004). An event history analysis of recurrent child maltreatment reports in Florida. Child Abuse & Neglect, 28(9), 947–966. [DOI] [PubMed] [Google Scholar]
  56. Loman LA & Siegel GL (2015). Alternative response in Maryland program evaluation. Final report to the Department of Human Resources, Social Security Administration. St. Louis, MO: Institute of Applied Research. [Google Scholar]
  57. Loman LA & Siegel GL (2013). Hughes et al: Science or promotion? Research on Social Work Practice, 23(5), 554–559. [Google Scholar]
  58. Loman LA, & Siegel GL (2012). Ohio alternative response evaluation extension interim report. St. Louis, MO: Institute of Applied Research. [Google Scholar]
  59. Loman LA, Filonow CS, & Siegel G. (2010). Ohio alternative response evaluation: Final report. St. Louis, MO: Institute of Applied Research. [Google Scholar]
  60. Loman LA,& Siegel GL(2004). Differential response in Missouri after five years: Final report. St. Louis, MO: Institute of Applied Research. [Google Scholar]
  61. Marshall SK, Charles G, Kendrick K, & Pakalniskiene V. (2009). Comparing differential responses within child protective services: a longitudinal examination. Child Welfare, 89(3), 57–77. [PubMed] [Google Scholar]
  62. Marshall DB & English DJ (1999). Survival analysis of risk factors for recidivism in child abuse and neglect. Child Maltreatment, 4(4), 287–296. [Google Scholar]
  63. Maryland Department of Human Resources (2020). Maryland’s Family Risk Assessment (MFRA), SSA# 04–03. Retrieved from: https://dhs.maryland.gov/documents/SSA%20Policy%20Directives/Child%20Welfare/SSA%2004-03%20MFRA%20Policy%20Directive.pdf [Google Scholar]
  64. Maryland Department of Human Resources (2015). Alternative response, SSA-CW #15–19. Retrieved from: https://dhs.maryland.gov/documents/SSA%20Policy%20Directives/Child%20Welfare/SSA%2015-19%20%20Alternative%20Response.pdf
  65. McCallum K, & Cheng AL (2016). Community factors in differential responses of child protective services. Public Health Nursing, 33(2), 107–117. [DOI] [PubMed] [Google Scholar]
  66. Merkel-Holguin L, & Bross DC (2015). Commentary: Taking a deep breath before reflecting on differential response. Child Abuse & Neglect, 39, 1–6. [DOI] [PubMed] [Google Scholar]
  67. Merkel-Holguin L, Schene P, Connolly M, Yuan YYT, Shusterman GR, Fluke JD, ... & Astolfi C. (2005). Differential response: A common sense reform in child welfare. American Humane Association Protecting Children, 20, 2–7. [Google Scholar]
  68. Montalvo-Barbot A. (2012). Some correlates of crime in eighty-one low-population-density and predominantly white counties in Kansas. Emporia State Research Studies, 41, 67–86. [Google Scholar]
  69. Maguire-Jack K, & Font SA (2014). Predicting recurrent maltreatment among high-risk families: Applying the Decision-Making Ecology Framework. Children and Youth Services Review, 43, 29–39. [Google Scholar]
  70. National Conference on State Legislatures. (2019, December 15). Differential Response in Child Protective Services: Analysis of State Legislative Provisions. https://www.ncsl.org/research/human-services/state-legislation-differential-response.aspx [Google Scholar]
  71. National Quality Improvement Center on Differential Response in Child Protective Services. (2014). Final report: QIC-DR cross-site evaluation. www.differentialresponseqic.org [Google Scholar]
  72. Palusci VJ (2011). Risk factors and services for child maltreatment among infants and young children. Children and Youth Services Review, 33(8), 1374–1382. [Google Scholar]
  73. Piper KA (2017). Differential response in child protection: How much is too much?. Children and Youth Services Review, 82, 69–80. [Google Scholar]
  74. Piper KA, Rycus JJS, Kelly C, & Holzrichter LJ (2019). Issues In Differential Response: Revisited. Columbus, Ohio: The Center for Child Policy. http://centerforchildpolicy.org/assets/IssuesInDifferentialResponseRevisited.pdf . [Google Scholar]
  75. Proctor LJ, Aarons GA, Dubowitz H, English DJ, Lewis T, Thompson R, et al. (2012). Trajectories of maltreatment re-reports from ages 4 to 12: Evidence for persistent risk after early exposure. Child Maltreatment, 17(3), 207–217. [DOI] [PMC free article] [PubMed] [Google Scholar]
  76. Putnam-Hornstein E. (2013). Hierarchical linear modeling: Applications to social work. Journal of Social Work, 13(6), 599–615. [Google Scholar]
  77. Putnam-Hornstein E, & Needell B. (2011). Predictors of child protective service contact between birth and age five: An examination of California’s 2002 birth cohort. Children and Youth Services Review, 33(8), 1337–1344. [Google Scholar]
  78. Rivaux SL, James J, Wittenstrom K, Bauman D, Sheets J, Henry J, & Jeffries V. (2008). The intersection of race, poverty, and risk: Understanding the decision to provide services to clients and to remove children. Child Welfare, 87(2), 151–168. [PubMed] [Google Scholar]
  79. Russell JR, Kerwin C, & Halverson JL (2018). Is child protective services effective?. Children and Youth Services Review, 84, 185–192. [Google Scholar]
  80. Schene P. (2001). Meeting each family’s needs: Using differential response in reports of child abuse and neglect. In Best Practice, Next Practice, Spring, (pp. 1–14). National Child Welfare Resource Center for Family-Centered Practice. [Google Scholar]
  81. Sedlak AJ, Mettenburg J, Basena M, Petta I, McPherson K, Greene A, and Li S. (2010). Fourth National Incidence Study of Child Abuse and Neglect (NIS–4): Report to Congress, Executive Summary. Washington, DC: U.S. Department of Health and Human Services, Administration for Children and Families. [Google Scholar]
  82. Shaw TV (2006). Reentry into the foster care system after reunification. Children and Youth Services Review, 28, 1375–1390. [Google Scholar]
  83. Siegel GL, Filonow CS, & Loman LA (2010). Differential response in Nevada final evaluation report. St. Louis, MO: Institute of Applied Research. [Google Scholar]
  84. Simpson DG, Imrey PB, Geling O, & Butkus S. (2000). Statistical estimation of child abuse rates from administrative databases. Children and Youth Services Review, 22(11–12), 951–971. [Google Scholar]
  85. Slack KS., Berger LM, DuMont K, Yang M, Kim B, Ehrhard-Dietzel S, Holl JL (2011). Risk and protective factors for child neglect during early childhood: A cross-study comparison. Children and Youth Services Review, 33 , 1354–1363 [Google Scholar]
  86. Thieman AA, & Dail PW (1997). Predictors of Out‐of‐Home Placement in a Family Preservation Program: Are Welfare Recipients Particularly Vulnerable?. Policy Studies Journal, 25(1), 124–139. [Google Scholar]
  87. U.S. Department of Health & Human Services, Administration for Children and Families, Administration on Children, Youth and Families, Children’s Bureau. (2020). Child Maltreatment 2018. Available from https://www.acf.hhs.gov/cb/research-data-technology /statistics-research/child-maltreatment.
  88. van der Put CE, Assink M, & van Solinge NFB (2017). Predicting child maltreatment: A meta-analysis of the predictive validity of risk assessment instruments. Child Abuse & Neglect, 73, 71–88. [DOI] [PubMed] [Google Scholar]
  89. Waldfogel J. (2009). Prevention and the child protection system. Future of Children, 19(2), 195–210. [DOI] [PubMed] [Google Scholar]
  90. Walsh WA, & Mattingly MJ (Spring, 2012). Understanding child abuse in rural and urban America: Risk factors and maltreatment substantiation. (Issue Brief No. 50, Carsey Institute). Retrieved from http://scholars.unh.edu/cgi/viewcontent.cgi?article=1169&context=carsey
  91. Winokur M, Ellis R, Orsi R, Rogers J, Gabel G, Brenwald S, Holmquist-Johnson H, & Evans M. (2014). Program evaluation of the Colorado Consortium on Differential Response: Final report. Fort Collins, CO: Social Work Research Center, School of Social Work, Colorado State University. [Google Scholar]
  92. Wulczyn F. (2009). Epidemiological perspectives on maltreatment prevention. The Future of Children, 19(2), 39–66. [DOI] [PubMed] [Google Scholar]
  93. Yampolskaya S, & Banks SM (2006). An assessment of the extent of child maltreatment using administrative databases. Assessment, 13(3), 342–355. [DOI] [PubMed] [Google Scholar]
  94. Zhang S, Fuller T, & Nieto M. (2013). Didn’t we just see you? Time to recurrence among frequently encountered families in CPS. Children and Youth Services Review, 35(5), 883–889. [Google Scholar]

RESOURCES