Rethinking the Functioning of Child Protective Services

In articulating CPS functions, it is important to consider that in the United States within the social services sector, it is far from clear what is meant by concepts of harm to children. Further, other sectors (health, judicial) that have an impact on CPS functioning have divergent perspectives about what constitutes harm due in part to differences in their experiences (samples) with children and families, the roles these sectors play in civil society, and the nature of their organizational stakeholders.

This is aside from the related issue of the difficulty of the meaning of CPS substantiation (Drake 1996). While some efforts have been made to develop standardized methods to define abuse and neglect (Heyman and Smith Slep 2006), thus far these have not been operationalized in CPS systems. One consequence is that from a policy perspective the CPS approach is scattered. In 2011, rates of child maltreatment victimization, a proxy for harm based on substantiation decisions, ranged from 1 child to 23 children per thousand among 50 states, with a likely range of variability even larger at the county or sub-jurisdictional level (US Department of Health and Human Services, Administration for Children and Families, Administration on Children, Youth and Families, Children’s Bureau 2012). While the unknown level of underlying actual incidence of maltreatment may in fact vary from jurisdiction to jurisdiction, the variability in officially recognized maltreatment by CPS agencies most likely reflects heterogeneity in the implementation of decision-making policy (Baumann et al. 2011).

Among the related persistent issues for CPS is the definition of various forms of child maltreatment; generally speaking abuse, deprivation of necessities, sexual maltreatment, and psychological abuse. One thing we know is that estimates of maltreatment incidence and prevalence vary by the type of measurement we use and the source of measurement. There is a major gap in overall incidence and prevalence rates in comparing official statistics with self-report studies or NIS like studies (Sedlak et al. 2010). There are also major differences depending on the source of official data (e.g., Hospital ICD 10 vs. CPS) (Gilbert et al. 2012).

We have also not done a good job of clearly identifying relevant distinctions between key harm-related constructs including maltreatment, risk of harm, and severity of harm (Sedlak et al. 2010). A common concept is to consider risk of future reporting or future substantiation, and there appears to be some relationships between concrete indicators of severe harm with risk (Baird et al. 1993). Nevertheless, it is not clear whether the bulk of moderate to high risk cases is at all likely to experience harm or at least severe harm, and some sense that important indicators of harm, particularly for neglect, are likely to be distal (Widom et al. 2012). Except in a very broad and highly ambiguous sense as of yet, there are no consistent definitions that lend themselves to clearly informing policy and practice in addressing what is meant by harm. Perhaps, more importantly, we lack a shared understanding of harm that would lend clarity to improving outcomes for children and families, although the concept of trauma and the underlying genetic impacts may come closest (Mehta et al. 2013).

Unfortunately, when the layers are peeled back often, what CPS is left with is the responsibility for fairly rare but obviously severe harm events: fatalities, severe physical injuries, and extreme cases of neglect including confinement and starvation that afflict very young children. However clear this view is of CPS’ responsibility; attention focused in this direction rarely succeeds in improving the underlying severe harm event frequency; that is, changes directed at CPS systems on this basis appear to be unable to prevent similar events in the future. For example, the reported level of fatalities known to CPS has not changed much during the past 20 years (DHHS 2012). Of course, part of the reason is that the solutions for CPS systems scrutinized for their failures to protect children against fatalities and other low-frequency events typically result in blame being placed on decision-makers, the introduction of overly prescriptive policy, commensurate with a lack of attention to conditions in the system that make errors likely (Munro 2011).

Assessments or assessment processes are a key part of the decision-making process, and for many professionals and administrators in CPS, these are the first and last method for improving decision-making. Many formal assessments are designed to support decisions along the CPS/CW continuum. For example, the National Resource Center for Child Protective Services (NRCCPS 2010) has documented many of the risk and safety instruments currently in use in states and some of the policy. However, this line of research has languished, and in fact, compared to the 1990s, there is very little published research. Almost no innovative research in the way of new approaches to CPS assessment and decision support has been published, although some important development without peer-reviewed research has occurred (Turnell and Edwards 1999).

To the extent we know how well these tools and processes function, the evidence suggests that they are modestly effective at prediction and probably work better under controlled evaluation conditions compared to actual practice (Camasso and Jagannathan 2012; Shlonsky and Wanger 2005). Still, they are better than nothing, and the research is sufficient to suggest that they should be used. However, there are fundamental research questions about how they should be used and whether actual decision-making can be improved upon. For some time these questions have not been pursued, and one is left with the impression that we have solved the problem.

Assessments are important but have important limitations in supporting decision-making in CPS. Mostly this is due to contextualizing issues as illustrated by the Decision Making Ecology/General Assessment and Decision Making Model (DME/GADM) concept (Baumann et al. 2011). A key concept is the action threshold embedded in the GADM, separating assessment from action. In several studies, factors other than case characteristics have been found to influence decisions. The function of context and decision-maker characteristics at each point along the decision continuum is the focal point for determining how decisions are impacted by factors other than case assessment. This has been instructive in looking at service and placement decisions, thus far. There is also fertile ground for exploring issues of disparities, but this is ultimately limited by our understanding of decision errors (see below).

Decisions are actually made by a number of agents in CPS. While caseworkers are obvious and important decision-making agents, there are potentially many others depending on the system and the point on the decision continuum. In addition to the other agents such as supervisors and judges, an important area of decision-making work involves decisions in groups. These can be professional groups (e.g., team decision-making and red team models) or some combination of professional and family groups (e.g., family group conferences). There is some evidence that decisions involving groups are effective, but many of the same DME factors that influence individual decision-makers also operate with groups. Each decision-maker brings their own history and understanding of the system into the process and evaluates the likely perceived consequences in the context of their values (Baumann et al. 2011). Studies of how the range of decision-makers including groups affects decisions in CPS are needed in order to take these effects into account and devise better workforce and workforce development policies.

Decision-making errors are present in CPS decision-making and at the worst involve situations of severe avoidable maltreatment or even fatalities. Balancing these errors are service decisions that may have adverse perhaps distal (longer-term) impacts on children’s outcomes, including the possibility that placements may play a role in these outcomes (Doyle 2008). Without exception, decision-making in child welfare occurs under conditions of risk and most often under conditions of uncertainty (Wakker 2010). A major conundrum for CPS administrators, and that ties into our confusion about concepts of child maltreatment related harm, is to clarify what type of error their system would least like to make and then design policy and resources accordingly to reinforce this perspective.

Unfortunately, and regardless of values perspectives, there is an absence of information that would allow us to objectively determine which type of error valuing is best. This condition is called decision-making under uncertainty. This is in part because, despite our recognition of the presence of these errors, we have no idea how often they are made or broadly speaking why. As a consequence, we have only a very limited capacity to avoid errors. It would be naïve to assume that these errors occur due solely to a specific factor like judgment, training, resources, or policy although these all play a role. Nevertheless, the lack of knowledge regarding the commission of errors and the interplay of these many factors hinders the development of feedback mechanisms that might allow for better pattern recognition and valid expertise.

James Mansell and colleagues (2011) have had some success in defining errors for CPS decision-making at intake: the decision to respond to a referral with an assessment/investigation. Their study attempts to define the problem as one of decision-making under risk, rather than uncertainty which is more typical. They found that, in the presence of a major increase in the proportion of cases investigated, some children that should have been investigated/assessed are still missed. They were able to specify a response rate balance between the likelihood that an agency will fail to act when they should have or exceed the available financial resources by responding more often than they should have. Given the ability to quantify the balance point, the agency is able to articulate the likely degree of defined error going forward. One of the findings of their study is that the agency actually did a fairly good job of distinguishing cases that required an investigation at the intake point.¹

The difficulty and challenges in researching similar errors for decisions related to other points along the decision-making continuum, and in particular removals, are well recognized. There is no gold standard that will tell us that we should or should not remove a child.

To summarize, there is an absence of clarity around the expectations of the CPS system, and if we acknowledge that a basic role of CPS is decision-making, the lack of clarity is compounded because we have no real method to evaluate error. So, in answer to the question posed above about what research has to say about the functioning of the CPS system, it has as yet little to offer regarding whether CPS is functioning well or poorly overall.

The continuum of CPS decision-making is a systems construct, and it classically defines the structure of the response pattern in the form of a decisional state space (Wakker 2010). That is, the actionable decisions are defined consistently as a series of choices: the state space. In the CPS state space for each individual family or child, an evaluation of the consequences of the decisions at each point is made. Some decisions, like intake decisions with immediately observable outcomes, appear to have a greater decisional consensus (narrower range of variability) compared to decisions where the consequences are less well understood, and it may be more straightforward to transform these to decision-making under risk, where outcome probabilities can be determined. Unfortunately, decision-making under conditions of risk in CPS is rare (Mansell et al. 2011). For the most part, conditions of decision-making under uncertainty prevail, meaning that the probability of decision outcomes is objectively difficult to specify even after the decision is made.

So perhaps another way to approach the problem of decision-making under uncertainty in CPS is to modify the state space for decisions, that is, modify the available choices in the continuum. Triage has been discussed as an important aspect of CPS systems that may permit different configurations of decisions. One approach to triage is differential response which in some jurisdictions has emerged as a formal response to maltreatment (Merkel-Holguin et al. 2006). Another approach might be to distinguish different classes of familial conditions (Casillas and Fluke 2014; Trocme, 2013, personal communication). For example, Trocmé (2013, personal communication) has suggested that maltreatment reports could be classified as acute and chronic, and maltreatment may be a secondary concern. Drake et al. (2009) have argued that the substantiation decision is superfluous and should be discarded in favor of identifying reported maltreatment cases as open or not for ongoing services. These approaches represent different formulations of child maltreatment decisional state space in as much as they focus attention on attaining information where the focus is on taking actions. To be clear, a decision to substantiate may permit or inhibit taking actions in some jurisdictions, but there is no information content from the standpoint of what action to take.

One possible advantage of orienting the system differently is that classes of families and children could be more clearly aligned with their likely proximal and distal outcomes. These outcomes would necessarily encompass both safety and well-being. Given this perspective, one way of thinking about an assessment process is to identify the possible outcomes based on classification and create a more fine grained and distally integrated view of the possible decisional consequences for taking actions or not. This idea while beyond our current base of research is described from a conceptual perspective in more detail in the following sections.

Needs among children in the child welfare system can be defined as cognitive and socioemotional conditions relating to emotional/behavioral disturbances, speech/language impairments, and learning/developmental delays. These problems are particularly important as they occur at a much higher rate compared with children in the general population (USDHHS-ACF 2007a). Child welfare studies estimate that 23–80 % of children exhibit mental health problems, while chronic health problems range from 35 to 80 %, and educational difficulties range between 31 and 67 % (Chernoff et al. 1994; Clausen et al. 1998; Pilowsky 1995; Schor 1982; Simms 1989; Stein et al. 1996; Szilagyi 1998). Using a representative US sample, NSCAW showed that within 3 years of a child welfare investigation 22 % of children have been identified as having a learning disability (vs. 8 % of US children), 15 % with an emotional disturbance (vs. 6 %), and 13 % with a speech impairment (vs. 5–9 %; USDHHS-ACF 2007a). Similar numbers have been found in Australia, Sweden, and Canada (Bromfield et al. 2010; Hessle and Vinnerljung 1999; Trocmé et al. 2010). Compounding these problems, it appears to be the norm that children with a special need tend to have multiple needs. In fact, 42 % of children in this US National Survey of Child and Adolescent Well-Being (NSCAW) who were identified as having a special need actually had two identified needs, and another 31 % had three or more needs (USDHHS-ACF 2007a).

Likewise, caregivers exhibit a number of service needs, including substance abuse, mental health problems, and cognitive impairments. Some of these needs are especially pervasive. For example, in one study of families involved in a child welfare investigation, Burns and colleagues found that 40 % of mothers suffered from depression (Burns et al. 2009). The mental health needs of caregivers of young children (ages 1 to 4 years) in the NSCAW seem to be similar, with 46 % of caregivers experiencing major depression at some point across the 5- to 6-year wave 5 follow-up (USDHHS-ACF 2009). Furthermore, 40 % of caregivers of young children had major depression at more than one point in time. Likewise, caseworkers’ reports on families with an infant or adolescent show that caregivers abuse alcohol (7–14 %) and drugs (8–30 %) and have serious mental health problems (18–23 %), cognitive impairments (7–11 %), poor parenting skills (41–43 %), and unrealistic expectations of their child (15–22 %; USDHHS-ACF 2008a, b). As with children followed in the same study, these problems tend to co-occur (USDHHS-ACF 2009). Other co-occurring needs include basic living assistance (33 %, e.g., transportation, food, housing), child care assistance (30 %), individual services (19 %, e.g., job-related, support groups, legal aid), and home assistance services (12 %, e.g., home management training; USDHHS-ACF 2008a). While these issues come from US child welfare data, similar challenges are also of concern in other nations such as Australia (Bromfield et al. 2010), Canada (Burnside 2012), England (Cleaver et al. 2007), and Scotland (Rosengard et al. 2007).

Unfortunately, it appears that both child and family needs are being underserviced in the current child welfare context. With respect to children, findings from the NSCAW indicate that children investigated for child maltreatment in the United States have a much greater risk of cognitive and socioemotional problems when compared with normed national samples (USDHHS-ACF 2005). However, they are not receiving needed services. Depending on the time of assessment, between 35 and 42 % of children assessed at baseline when they were 3 years of age or younger had a developmental delay or an established medical condition or will have developed one within the next 5–6 years. Yet only 10–26 % of those in need received necessary developmental or medical services, with percentages worsening the longer children were involved with the child welfare system (USDHHS-ACF 2007b). Similarly, although 31 % of preschoolers had cognitive deficits and 27 % had behavior problems, only 13 % and 12 %, respectively, of those in need received special educational services to address them (USDHHS-ACF 2005). The situation is similar for school-age children investigated for maltreatment in the United States. Depending on the measure, 5–12 % of school-age children exhibited cognitive development or academic achievement risks (intelligence, reading, math), and another 10–45 % exhibited developmental risk on one or more socioemotional measures (problem behavior, social skills, living skills, depression). Yet, only 54 % of those school-age children who exhibited risk on at least one of the measures were receiving special education services for cognitive/academic difficulties and 25 % for socioemotional functioning (USDHHS-ACF 2005).

With respect to caregivers, although 27 % had a mental health need, a minimum of 26 % of those in need received a mental health service (USDHHS-ACF 2008a). Most often the mental health service received was psychotropic medication, with few actually ever seeing a mental health practitioner. Similarly, almost no caregivers of infants received substance abuse services (though 30 % were in need). The services more often provided were basic living assistance (33 %, e.g., transportation, food, housing), child care assistance (30 %), individual services (19 %, e.g., job-related, support groups, legal aid), and home assistance services (12 %, e.g., home management training; USDHHS-ACF 2008a).

Although far fewer children and caregivers receive services than those in need, involvement with child welfare has been shown to increase service receipt. That is, similar children and families who are not involved with the child welfare system are even less likely to receive needed services. For instance, across child age groups, there is an increase in receipt of school-based or specialty mental health services for child welfare cases (Leslie et al. 2005). Thus, despite high levels of unmet need, the ability of the child welfare system to connect families to services is promising.

However, consistent with the DME framework (Baumann et al. 2011), factors other than the type of services needed appear to influence which services are received. For example, for children 3 years and under, the percentage of those with developmental or medical risks receiving services does not differ from the percentage of those without such risk receiving services. Furthermore, although children in substantiated and unsubstantiated cases have similar special needs (Drake 1996; USDHHS-ACF 2007c), service receipt is instead best predicted by substantiation and poverty, irrespective of current developmental status (at least between baseline and 12-month follow-up for children 3 years and under; USDHHS-ACF 2007b). This is important because services are being implemented similarly across clients regardless of their actual need, resulting in wasted service provision for some, and lack of service provision for others that actually need it. Such inadvertent care translates into financial waste or potential deterioration in functioning depending on which group is in question. The child welfare field must better understand these complex needs that touch multiple systems and better integrate across systems to get those needs addressed.