Mechanisms of Inequity: The Impact of Instrumental Biases in the Child Protection System

Abstract

The structural risk perspective conceptualizes the causes of inequities in child protection system contact as unequal exposure to the structural causes of child abuse risk, combined with biases in the responses of child welfare workers and reporters. This conceptual article proposes a third mechanism of inequity: instrumental biases. It is proposed that instrumental biases operate as a third group of mechanisms that inequitably increase the involvement of some groups and not others. Instrumental biases operate through institutional structures, interpretive concepts and risk proxies that affect how risk is coded and becomes attached to particular people. Against the background of the notify-investigate model that creates poor conditions for decision making, and shapes institutional structures, instrumental biases include the miscalibration of the demand and supply of services (an institutional cause); family-specific surveillance bias and a reliance on prior case histories (a risk proxy cause); widening legal definitions of serious harm (an interpretive concept cause); and complex responses to intimate partner violence that minimize theories of IPV and the social context it occurs within (concept and risk proxy causes). It is argued that within the decision-making context of the child protection system, how services are structured and risk becomes codified has disproportionate impacts on some communities compared to others. Examples from Aotearoa New Zealand, with reference to Māori and people living in high-deprivation areas, are used to illustrate these concepts.

1. Introduction

Despite the best intentions of child protection systems in many countries, intersecting ethnic and socioeconomic inequalities are persistently reflected and, in some cases, intensified by them. Structural inequalities are systematically related to chances, experiences and outcomes within the child protection system [1]. Yet the mechanisms by which these inequalities manifest in systems are not well understood, and have contextual variations and many nuances. Understanding the mechanisms of how inequalities become embedded in child protection systems, as well as exceptions and contingencies, is essential to meaningful responses [2,3,4,5]. There is a growing consensus that efforts to reduce both disproportionality and disparities must address mechanisms both inside and outside of the statutory child protection system. External mechanisms cause increases to the risk of harm, and internal ones cause biases that escalate some people into system contact while reducing the chances of others [3,4,6,7,8].

Drawing on publicly available data and published research, this article develops a conceptual framework that extends the risk/bias (or structural risk) approach to include instrumental biases. It first defines this concept as biases that operate via institutional structures, interpretive concepts and risk proxies, then situates it within the existing risk/bias theoretical model. Drawing on inequalities and decision-making theories, several mechanisms that are examples of instrumental biases are then proposed. Examples are drawn from the literature that fit the definition, with a focus on Aotearoa New Zealand. Those selected are: demand and supply of services, family-specific surveillance bias and prior contacts, widening legislative definitions, and responses to intimate partner violence within the child protection system. The implications are that inequities are driven by factors beyond increased risk and direct biases; therefore, complex responses that take account of these are needed. With relevance for other countries where ethnic, racialized and socioeconomic inequities occur, this article tentatively describes the role of instrumental biases and discusses the implications for policy.

2. Inequalities in the Global and Aotearoa New Zealand Contexts

Inequalities in the chances of contact with child protection systems, and the experiences and outcomes of these systems, is a global issue, including in the UK, the US, Australia and Canada [6,9,10,11,12]. Wider social inequities relating to gender, class, ethnicity, location and disability are profoundly reflected and refracted by contact with that system, sometimes in unexpected ways [13]. The processes that lead to these inequalities are still being explored, and differ between national and regional contexts. System reform researchers and advocates have long drawn attention to the inadequacies of system structures that are focussed on risk, individuals, and forensic responses to child abuse and neglect, instead outlining the need for coherent, public health type approaches to child abuse prevention focussed on reducing known risk factors present in the social context [14,15]. Systems established to respond to risk of abuse and neglect are often ill-equipped to provide poverty alleviation to address family material needs, to address historic inequities and cultural denigration for Indigenous peoples, or provide therapeutic services that respond to the trauma and distress experienced by families that can lead to risk of harm for children.

Instead, based on a centralized ‘notify-investigate’ model, they tend to become risk focussed and this provides fertile ground for the reproduction of biases, both social and cognitive [16,17]. This is because the notify-investigate institutional structure (a central agency that others ‘notify’ families to, that then ‘investigates’ them) invites people to make decisions that draw on vague definitions, poor information quality, contested definitions and short timeframes, while providing feedback that exacerbates base rate fallacies and other cognitive biases [16,18]. At the same time, without addressing the social conditions that disproportionally affect Indigenous and racialized minorities, the structural determinants of contact remain. The interface between structural determinants, and biased decision-making seem to be exacerbated by several instrumental biases that affect the ability of the system to address disproportionality effectively.

Descriptive correlation studies that map out inequalities provide the basis of understanding how inequities are patterned (though they struggle to explain them). Many, for example, have outlined the correlative relationships between deprivation, low income, poverty, or inequality in the chances of system contact [13,19,20]. These studies show the steep social gradient related to poverty or inequality. In Aotearoa New Zealand, this gradient is substantial. Children living in the most deprived decile are 21 times more likely to be substantiated for child abuse and neglect, 18 times more likely to have a Family Group Conference (a legislated family meeting to compose a plan to address the child protection concerns) and 9.4 times more likely to be in care than those in the least deprived decile of neighborhoods [5].

Māori (the Indigenous people groups of Aotearoa New Zealand) disparities are also significant in the child protection system. Seventy percent of children in care are Māori, despite Māori children being approximately 25% of the child population [21]. Māori disparities have a significant intersection with deprivation levels (as defined by the NZDep index). The Māori rate of placement into care is 851 per 100,000 in the most deprived quintile, compared to 350 per 100,000 in the least deprived quintile [22].

Pākehā (people descended from Europeans) rates also intersect with deprivation to produce inequalities. The Pākehā rate in the least deprived decile is 78 per 100,000, but 616 per 100,000 in the most deprived decile. The social gradient is therefore flatter for Māori (2.4 times the difference between low/high deprivation) than for Pākehā (7.9 times). The most extreme differences are found when comparing placement rates for Pākeha in low-deprivation areas with Māori in high-deprivation areas, which is 78 per 100,000 to 851 per 100,000, a difference of 10.9 times [22].

The number of children entering care in Aotearoa New Zealand has reduced sharply in recent years, and the proportion of Māori children entering care has also reduced, halving in raw numbers from 2017 to 2020, as seen in Figure 1. Māori care entries have also reduced as a proportion from 67% of care entries to 58% [23] (Māori and Māori-Pacific children summed for total Māori). The overall numbers of children in care (not entries), however, has only reduced slightly, from 6316 in 2017 to 5945 in 2020. The ethnic breakdown of children in care is not known (as opposed to care entries), as it is not reported in current available data. This reduction in care entries, not able to be discussed fully here, has been linked to several public inquiries and increasing calls to decolonize the child welfare system (see [24]).

3. Risk, Bias and Instrumental Biases

Drake’s risk-bias model and related versions of it have been used to tease out the questions relating to the mechanisms causing these kinds of inequalities [7]. They ask: are some populations over-exposed to risk factors for child abuse and neglect (such as poverty, discrimination, and colonization), which increase the incidence of child abuse itself; or are heightened rates of system contact related to exposure and surveillance biases outside the child protection system, and the direct bias of caseworkers once they enter it? [25,26,27,28]. Both national and international research frame disparities experienced by Indigenous, and in some countries, particular racialized ethnic minorities, as related to this ‘risk-bias’ or ‘need-bias’ debate [7,8,25,29,30].

Over-exposure to poverty of some groups may increase the true incidence of child abuse and neglect by placing stress on family life and reducing access to remedial resources, and the damaging effects of structural racism and colonization may impact parenting capacity and extended family capabilities [31,32]. Stressors are exacerbated by contexts where there are few universal social protections, low community social cohesion, and poorly matched quality and quantity of culturally-relevant services [33,34,35]. This direct increase in risk combines with a number of biases to produce disparities. These biases include the direct bias of practitioners (both notifiers and child protection workers), the over-surveillance of some groups, and visibility bias [30,36]. Combining these perspectives in a both/and model is recognized as a realistic perspective based on existing research, and this has recently been coined a ‘structural risk perspective’ [4].

A structural risk perspective incorporates both risk and bias explanations. Feely and Bosk [4] define a structural risk perspective as one that “builds on both the bias and differential risk perspectives by explicitly considering the role that structural socioeconomic conditions play in shaping unequal CPS involvement while also acknowledging individual explicit and implicit bias…disproportionality emerges from the structural racism which shapes our society, which results in the unequal distribution of resources and opportunities, and which elevates the risk of maltreatment risk … these inequalities are further amplified by biased decision-making that occurs throughout the system…the structural risk explanation views a fair response to racial disproportionality as appropriately identifying all the places where maltreatment risk is situated…(as well as) continued intervention related to individual bias” (p. 4). This article builds on that perspective to describe a theoretical model that adds instrumental biases as a further mechanism perpetuating inequalities.

In most countries, including Aotearoa New Zealand, both risk and bias contribute to disparities in child protection system contact for those living at the intersections of high-deprivation areas and those who are Māori (Indigenous people of Aotearoa New Zealand) [5,25,30]. To state is it one and not the other tends to reduce the scope of responses and leads to finger pointing and disavowals of responsibility [3]. Instead, a both/and approach to disparities is needed in order to devise a reasoned strategy to respond to both disproportionate need within some Māori whānau (extended families) and families in high-deprivation areas as well as address biases within the systems that respond. Both risk and bias, as sources of disparities, can be related to patterns of racism, colonization, and class inequity through history, rooted in both cultural and economic imperialism [17,37,38].

Structural risk highlighted in the US context chimes with both historical documents and current inquiries in Aotearoa New Zealand. For example, the ‘Puao-te-ata-tu, It is day break’ report [39] (which emanated from an investigation into the child protection system and disproportionality for Māori children at that time) describes the institutionalised racism affecting Māori communities, decision making and reporting. It notes the contested cultural conceptual underpinnings of concepts as basic as family, childhood and children. More recent reports such as the Whānau Ora and Office of the Children’s Commissioner reports also point to the systemic effects of racism, embedded in colonialism, on Māori communities, as well as decisions within the child protection system up to the present time [25,38].

The Waitangi Tribunal came to a similar ‘both/and’ conclusion in Aotearoa New Zealand after investigating the causes of disparities for Māori, noting that these relate to factors both external to Oranga Tamariki (the child protection agency in Aotearoa New Zealand) and internal [25]. The external factors result from colonization that led to the social conditions of poverty, cultural denigration and racism, particularly “racism, historical injustice, intergenerational trauma and persistent inequity across a range of social wellbeing and socioeconomic measures” [25] (p. 51). Poverty and its connection to raupatu (land confiscation) that forced urbanisation is also outlined: “Poverty has a huge impact on Māori whanau and their ability to look after their children. The fact that many Māori whanau live in poverty and are divorced from their papakainga or shared collective whenua is a direct result of raupatu and colonization. It is no wonder we are seeing flow-on effects like homelessness, other socioeconomic issues, and so forth” (Jakeman, in [25] (p. 51)).

The internal factors include the continuing devaluing of Māori knowledge systems, particularly understandings of children and their connections with whānau that differ from Pakeha constructs. They note that: “the Maāori child should not be viewed in isolation, or even as part of a nuclear family, but as a member of a wider kin group or hapuū community that has traditionally exercised responsibility for the child’s care and placement … through whakapapa, tamariki are endowed with attributes fundamental to their cultural, physical, and spiritual well-being such as mana, tapu, wairua, and mauri. Further and importantly, as Mr Shortland notes, rangatiratanga is the inherent birthright of all tamariki Maāori” [25] (pp. 60–62). These examples highlight the context-specific interplay of risk and bias factors that perpetuate inequalities in Aotearoa New Zealand.

4. Defining Instrumental Biases with Examples

Beyond these established bias and risk factors, the role of instrumental biases have been less explored. Instrumental biases are factors that embed institutionalised inequities via neither the direct bias of individuals, nor increased risk of child abuse and neglect, though they may contribute to their pre-conditions. Instrumental biases can be defined as: mechanisms that are not related to either increased risk or direct bias, but operate via third-party mechanisms that have their genesis in institutional structures and processes, interpretive concepts, and risk proxies (drawing on [16]).

‘Institutional structures’ include the notify-investigate model of child protection systems. This model consists of a central body that holds most statutory power, and various community organisations who have some responsibility to report child abuse to it. These are connected through a web of contractual obligations and regulatory guidance that creates reporting obligations and differential service responses [16,17]. This model exacerbates bias potential by requiring reports from community reporters, as well as decisions made at the point of intake, under the poorest of conditions for decision making, that is sparse information, historic information, time pressures and limited relational knowledge [17]. Reports may also be shaped by surveillance bias. While the biases are direct, they are supported by the instrumental bias created by this institutional structure. The demand and supply of services within this notify-investigate model, discussed below, can also affect inequities in addition to the basic underpinning structure, by reducing the provision of culturally relevant services or creating ‘supply-suck’ [40].

‘Institutional processes’ are processes such as decision-making fora, screening processes, the use of particular technologies, assessment tools (these also contain specific concepts) and workplace conditions. ‘Interpretive concepts’ include formal knowledge bases applied to understand family behaviour as well as dominant institutional discourses or heuristic logics (‘this kind of case gets this kind of response’), combined with the effects of the subjective social location and values of practitioners [41]. Interpretive concepts also include the formal theories espoused by either organizational guidance or the professional education involved (usually of social work) to guide understandings of service user presentation, which may be culturally contested (such as trauma and attachment theory) [41,42,43]. In addition, interpretive concepts include legal definitions as they are codified and applied in practice.

‘Risk proxies’ occur when there is a reliance on certain superficial factors to denote risk or legal thresholds, that have a disproportionate impact on particular communities (such as perceptions of compliance with authorities in contexts where there is low historic trust in public bodies and services, or multiple prior reports) [44]. When enacted via practitioner judgement, risk proxies can intertwine with cognitive heuristics and biases to categorise people as high or low risk, leading to greater intervention for people from some communities, and lower for others. These proxies do not necessarily reflect incidence of abuse, but become relied upon in decision making as if they do. This article now turns to give further detail regarding selected examples of instrumental biases.

5. Demand and Supply of Services as an Instrumental Bias

As mentioned, an instrumental bias related to ‘institutional structures and processes’ that may contribute to inequities in system contact is the demand and supply of services, and particularly the balance between social protections, therapeutic support services, intensive family preservation and child protection services [45,46]. The over-supply of child welfare services in any given location can result in ‘supply induced demand’, where the supply of child protection services pulls people into the system who may not need it. This can particularly affect those from racialized groups [6,40,47,48]. On the other hand, too much ‘demand’, that is overwhelming notifications, can reduce the proportion of cases ‘screened in’ due to the system being overwhelmed, and, disparities can also be reflected in the decisions relating to screen-in for a range of reasons [49,50]. One is the supply of preventive services outside the statutory service, as their quantity, quality, and cultural acceptability affects patterns of contact with the statutory system, including inequalities within those patterns. For example, an increase in preventive services can reduce entries to care of children living in high-deprivation area [51].

A key question is: if the problem is over-supply, or undersupply, of either statutory or preventive oriented services, how do these dynamics interact with disparities relating to class or ethnicity? Does a high risk threshold (resulting from undersupply of statutory services) reduce disparities or increase them, and why? Wulczyn [40] found that the contextual county level variable of ‘supply signal’ (high supply of congregate care) and the presence of a standardised assessment both affected disparities, finding that “if there is no supply signal, driving congregate placement rates higher, and no mandatory assessment tending to push congregate care rate lower, the Black/White difference grows, and White/Hispanic differences shrink” (p. 23). As the authors note, the nature of services supplied not only creates flows in and out of the system due to material provision, but service type and availability constrain the decisions available to workers.

For example, Choi, Kim, Roper, LaBrenz, and Boyd [52] found that children of colour, and from poor backgrounds, were less likely to be assigned to the ‘differential’ pathway at intake, providing less opportunity for support services to assist. This alone increases inequities in child protection system contact, but may not be simply class and racial bias of intake workers. Fallon et al. [49] found that the disparities experienced by Indigenous Canadians were correlated (after regression) with the proportion of Indigenous people living in a particular area, and the available provision of culturally appropriate preventive services, rather than Indigenous identity alone. This finding suggests that the availability of services of the right type affects the range of options available to decision makers regarding Indigenous families, and this affects disparities. If a practitioner knows there is no culturally appropriate support service available, they may be more likely to screen a case in. This is not the same as direct bias that equates Indigenous or minority status with risk; it is more that the perception that a specific culturally responsive type of service is not available that leads to a decision based on concern the risk cannot be effectively addressed elsewhere.

Differences related to either location-specific demand and supply of services or site-specific practices have been further explicated by studies that found an ‘inverse intervention’ rule. These studies found, in contexts as diverse as England, Aotearoa New Zealand and Texas, that an inverse intervention rule was operating. This observation finds that children living in highly deprived small areas, surrounded by less deprived larger areas, experience higher rates of intervention from child protection services than if they live in a highly deprived larger area [1,5]. Several reasons have been proposed including different organisational cultures that fast track children into care in less-deprived areas, or a greater resource of child protection services creating ‘supply suck’. Others propose that there are more prevention services provided in highly deprived areas that are keeping children out of the care system; or that there are not enough child protection services to intervene due to too much demand. The interplay therefore between the quantity, type and nature of service provision, both statutory and in the community, can act as an instrumental bias to either increase disparities or reduce them.

6. Risk Proxies, Family-Specific Surveillance and Instrumental Biases

Risk proxies are implicated in instrumental biases. Direct biases occur when assumptions about particular people based on race, class or gender affect risk perceptions. But instrumental bias can occur when known cognitive heuristics and biases intersect with the codification of risk that occurs within organizational cultures, creating risk proxies only tangentially related to actual risk of harm [53,54]. For example, family-specific surveillance bias may be another instrumental bias that establishes a risk proxy, that in turn increases inequities in system contact. General surveillance bias has been identified as one mechanism that can contribute to inequities [36]. Surveillance bias refers to the over-surveillance of particular communities and neighborhoods, particularly by professional referrers. In the US context, this is widely reported, though also contested [28,55,56,57]. While group or location-based surveillance may be one cause of surveillance bias, there are other mechanisms by which surveillance may function in more specific ways. For example, is surveillance equally experienced for all Māori and all people living in poorer areas, or a subset of these intersecting populations? Does it have nuanced interactions with a history of family contact and histories of colonization and racism?

A more nuanced mechanism may operate through the surveillance of specific extended families (or whānau). Keddell & Hyslop [30,58] found that practitioners often identified ‘famous families’ who lived in their area, and who had had extensive intergenerational contact with the local site office. For these families, who were often Māori, simply having a particular family name or known family connection could heighten the investigative response from Oranga Tamariki. As Māori historic overrepresentation is clearly linked to cultural and institutional racism, this perpetuates historic over-intervention in a pernicious feedback loop [38,39]. This means that while surveillance biases may affect whole population groups, they are exacerbated by assumptions about the meanings of prior system contact for a specific subset of Māori whānau (extended families).

Families with multiple contacts with multiple systems also appear to generate a heightened perception of risk, even if those services are aimed at assistance, support or health needs. Often accessed easily through information-sharing arrangements, these integrated data increase the perception of risk for high-need families. This may inadvertently convert high needs, into a high perception of risk for families for whom the social determinants of health disproportionately affect (Indigenous and racialised minority groups). This is a cause of concern for community-based social workers, who perceive this as unfairly heightening risk perceptions for some, mostly families living in high deprivation: “Often there’s a, like an overshadowing, I mean there’s a care experience part of things but an overshadowing of some connection with a justice system or a health system. It might not even be current …so whether it be family violence or there’s been drug and alcohol …The concerning thing for me is it doesn’t need to be a current thing but it’s, you know a past thing that’s been there and because that family is then under some sort of radar within the systems, they’re jumped up a level or two for people to react” (Practitioner B, Focus Group 1) [59].

These examples suggest that surveillance bias may not apply equally to all Māori or all people in high-deprivation areas, but may contain further nuances relating to previous system contact, specific known families, or contact with other systems. The history of colonization and the deep interconnections between poor health, criminal justice system contact and poverty mean this is particularly impactful for some Māori whānau. This is because conviction rates and sentences of incarceration are higher for people from Indigenous and other racialized groups than those from White groups due to histories of racism and or colonization [60]. This can contribute to confirmation biases, where those with convictions attract a filtering out of positive information and active searching for negative evidence. The individualisation of these risks in the child protection system shows the need to have a system that addresses the social causes of harm rather than codifies them as risk attached to the individual [61]. From an instrumental bias perspective, it may also continue to cycle some families through the child protection system based on superficial assumptions about risk, while missing others altogether.

A related bias is the use of recorded case history in assessments. An emphasis on ‘the history’, as contained in case records, at the expense of current, in-depth assessment based on an engaged relationship will exacerbate biases because of the weight of history—more Māori are surveilled and have intergenerational whānau histories of system contact. Giving weight to recorded contact in decision making without up-to-date assessment compounds historic inequities. This emphasis can be seen in the research of Keddell et al., [59], where their respondents noted that there was often an emphasis on ‘what was written on the paper’ instead of what family members are reporting about current circumstances. This was also noted in other research by the Children’s Commissioner, and in the Hawkes Bay case review, which found that: “New information about the mother and whānau that was shared with Oranga Tamariki but was not consistent with the historical information appears to have been discounted, overlooked or not pursued”, and “Historical concerns were relied upon to form the basis of decisions for the older child as justification for a set pathway for this child” [62] (n.p.).

As Feely and Bosk [4] note, a reliance on case histories or prior reports can exacerbate ethnic disparities because they will identify as high risk those families most vulnerable to socioeconomic hardships and the associated stressors on family life that may make previous reports more likely. In a colonized context, historical and current racism based on the assumed superiority of Pākeha middle-class parenting norms may drive up reports for Māori, including the reach of data available that can now show a parent’s own involvement as a child, and the involvement of their parents [41].

These factors can inadvertently bake in bias to the assessment of individuals, even when there are no direct racist attitudes held by the practitioner—they have simply internalised a set of risk codes or proxies about the meaning of historic contact that happens to disproportionally affect Māori and others living in high-deprivation areas [63]. These interpretive biases based on proxies for risk, are heightened within an institutional system that relies on notification to a centralized service that must then make decisions at intake under poor decision-making conditions as mentioned above—limited information, worker stress and strict timeframes, the precursors to superficial and biased judgements [18]. ‘Thin’ information in turn can exacerbate the conditions for bias, as a reliance on intuitive heuristics is likely to be higher when information quality is poor, as cognition tends to ‘fill in’ knowledge gaps to create coherent narratives [64,65,66]. This may mean that the presence of biases as a source of disparity may be stronger at intake decisions, but lower at other decision points due to increased information depth and greater consensus around the severity of the issues faced by the family [30].

7. Broadening Legislative Definitions of Serious Harm in the Context of Decision Making

Among types of instrumental bias in addition to risk proxies are the concepts used to interpret family behaviour and situation. Importantly for inequalities, it is clear that biases relating to ethnicity interrelate with cognitive biases such as confirmation bias and availability heuristics [67,68]. These types of biases are most pernicious when there is less consensus about the level of harm or definition of the situation as abuse, creating wide decision-making discretion [69]. In Aotearoa New Zealand, as in many countries, this vagueness around definitions of abuse is reflected in wide discretion in the governing legislation.

The legislation in Aotearoa, the Oranga Tamariki Act 1989 for example, has been widened even further in recent times, by changing the definition required for intervention. This is defined in section 14 as a child who ‘is suffering, or is likely to suffer, serious harm’. This must be applied, however, with regard to the new section 14AA, which took effect on July 1 2019. This new subsection provides a wide interpretation of serious harm or its likelihood, for example:

“(a) a child’s or young person’s development or physical or mental or emotional well-being is being, or is likely to be, impaired or neglected, and that impairment or neglect is, or is likely to be, avoidable:

(b) the child or young person has been exposed to family violence (within the meaning of section 9 of the Family Violence Act 2018)):

(3) For the purposes of applying section 14(1) (a) and subsections (1) and (2), serious harm may occur (without limitation) as a result of—

(a) an incident; or

(b) 2 or more incidents that taken on their own would not be serious enough to constitute serious harm, but the cumulative effect of which is serious enough to cause serious harm; or

(c) the co-existence of different circumstances”

[70].

These new subsections extend the definition of serious harm to include vague factors such as the future emotional well-being of a child, something that can only be guessed at and relies on subjective and contested interpretations of risk of future emotional harm [12,71]. It is also extremely difficult to disprove, particularly when research that shows the greater likelihood of harmful effects of specific types of abuse across a population is utilised. That is, research usually shows an increased risk of future poor outcomes, but this is often nuanced, shaped by factors such as age and gender, abuse severity and personal and social factors of the child. Such studies are probabilistic, that is, not definitive in their effects, and downplay the socioeconomic causes of future harm, yet are often used to shore up claims of absolute risk of future harm to all children exposed [72,73].

The new subsection also includes exposure to intimate partner violence specifically, without the exceptions noted in the Family Violence Act 2018, that the adult victim cannot be held liable for the harm of exposure [74]. As noted in the section on IPV, IPV can have varied effects on children; and in the child protection context, can exacerbate inequities due to the ways that IPV prevention can be assumed to be the responsibility of women survivors.

The addition of ‘two or more incidents’ that would not, on their own, have constituted harm is also now included, as is the ‘co-existence of difference circumstances’. Such vague and broad definitions of serious harm further extend the legislation and the lack of clarity about when its provisions can apply. This may set the scene for increasing bias, because a lack of definitional clarity, especially around key decision points, is a key aspect of highly subjective interpretations of harm, which can reflect social stereotypes and cognitive biases of many kinds. The extension of categories to widen legislative catchments has been described as ‘concept creep’, capturing more and more families in the CPS net, while doing little to address the causes of harm, including that caused by exposure to IPV [75,76]. Legal scholars note that vagueness in definitions escalates biases. Alcantara [77], for example, notes that increasingly vague patent definitions, as part of a wider move away from ‘bright line’ tests, may be disadvantaging female applicants.

Keddell & Hyslop [30] found that Māori children were perceived as more at risk than Pakeha children in the same situation; however, this bias was most evident when the case characteristics were in close proximity to the relevant threshold decision. Where the level of concern was low, for example, there was little difference shown in perception of risks between Māori and Pākehā. But when the case details approached decision points, the perceptions of reaching the formal decision point diverged, and bias became more evident. This may show that bias has the greatest effect when the case factors are decision proximal, that is, they are near to a particular threshold. It may also show divergence between risk perceptions and threshold decisions, that is, while risk perceptions show small differences, the perception related to whether the decision threshold had been reached showed much more difference between the ethnic groups, with the Māori whanau perceived as much closer to the decision threshold. Where there is a lack of surety or consensus about the nature, type and seriousness of abuse, biases have more effect. This may reflect differences between perceptions of risk and the threshold for taking action [3]. Both a legal and conceptual lack of clarity exacerbate these effects.

8. Intimate Partner Violence: Instrumental Bias or Increased Incidence?

One method of identifying instrumental biases is to examine major sources of reports to child protection systems, and explore if, and to what extent, those reports show heightened incidence, direct biases or some other instrumental mechanism that increases the escalation of particular groups into the child protection system, or represses escalation for others (drawing on Bywaters’ et al., 2015 definition of child welfare inequalities). One issue that can be examined as either a risk proxy or concept (as sources of instrumental biases) is intimate partner violence (IPV) [78]. This problem is the source of many reports to the child protection system in various countries, and usually enters the child protection system via police call outs [76,78]. Examining the relative contributions of risk, bias and instrumental biases can be applied to IPV in Aotearoa New Zealand. While there is no detailed statistical study comparing incidence with child protection outcomes, by comparing incidence with disproportionality and disparities in the child protection data, some tentative conclusions can be drawn about the causes of inequities via the IPV pathway.

In Aotearoa, there is an increased risk of IPV victimisation related to structural inequalities for Māori women and those in high-deprivation areas. Women living in the most deprived one-third of areas have a higher rate of self-reported prevalence of IPV victimisation both across their lifetime and in the last 12 months, at 35% compared to 23% in the least deprived one-third (in 2019) [79]. Only 13% of those who reported IPV in the Fanslow study had sought help from the police, court or lawyer. An earlier study found that only 12% had talked to the police about their experience of IPV, and regression showed that there was no difference by ethnicity or recentness of the violence as to who talked to the police, with the seriousness of the violence being the stronger predictor of seeking police help [80,81]. Māori are more likely to be victims of IPV (58%) compared to Pākehā women (34%) over their lifetime, with one study showing 80% of Māori women will be a victim of IPV and are 3 times more likely than non-Māori to be killed by a domestic partner [82].

Another study showed that after controlling for socioeconomic status, Māori had odds of IPV of 2.12 to 2.76 times those of non-Māori, and crucially, odds of 2.98 times non-Māori for injury [83]. It can reasonably be inferred that if Māori women are overrepresented in high-deprivation areas, and more likely to experience injury due to IPV, that is, experience more serious violence, then Māori women may also be disproportionately represented in the small proportion who call the police for help. This explanation would lead to a conclusion that increased reports to child protection were a function of increased incidence (risk) [80].

Calling the police can lead to a report to Oranga Tamariki, although there are a number of collaborative efforts to reduce this progression into the child protection system. There is also increasing reluctance for the child protection system to become involved unless the IPV is severe, inline with growing efforts to keep children within their families and access preventive options. There are a growing number of community services able to respond, and better coordination of these services between Oranga Tamariki and the police [84].

However, a closer examination of the data (from Official Information Act requests) tracking reports of concern from Police to Oranga Tamariki for intimate partner violence shows interesting patterns. Firstly, there has been a large reduction in cases accepted as reports of concern (ROCs) by Oranga Tamariki from the police for ‘family harm’ between 2010 and 2019, from 57,489 to 5,461. At the same time, police reports to Oranga Tamariki for other reasons have increased, from 17,777 to 22,529. Disparities for Māori for family harm reports, however, have increased over this time, from 52% of all reports in 2010 to 62% in 2021 [23]. A closer examination of rates, rather than just raw proportions, confirms this as can be seen in

Table 1. Rates of family harm (police) reports of concern to Oranga Tamariki for Māori and non-Maori, 2010–2019.

	2010	2019
Māori FHROCs	30,367 *	3389 *
Est. Māori pop U18	280,180 **	305,110 **
Population rate	0.1084	0.0111
Within population rate change 2010–2019	=9.7-fold reduction
Non-Māori, non Pacific FHROCs	13,775 *	1108 *
Est. NM/NP pop U18 ***	815,710 **	838,580 **
Population rate	0.0169	0.0013
Within population rate change 2010–2019	=13.0-fold reduction
Between population disparity rate	6.4×	8.5×

* From Official Information Act request data. ** From Stats NZ infoshare estimated resident populations. *** Non-Māori, non-Pacific population calculated by subtracting the estimated Māori pop from total.

. By calculating disproportionality ratios and comparing them, it can be seen that the Māori rate of ROCS for family harm declined by 9.7 times (970%) between 2010 and 2019. However, the non-Māori, non-Pacific rate declined at an even greater rate, by 13 times. This difference in reduction between groups on top of the existing disparity, means that while rates have declined immensely overall for all groups, disparities for Māori in reports of concern grew from 6.4 times that of non-Maori, non-Pacific to 8.5 times.

Comparing reports of concern for IPV to IPV cases then categorised as ‘further action required’ also shows a higher proportion of Māori at the latter of these two decision points every year since 2010 (Figure 2), despite the overall large reduction in reports. This small increase at this particular decision point serves to ratchet Māori whānau into more intervention while others take exit pathways.

Understanding the intersections between these patterns and: IPV severity, deprivation, and changing decision thresholds is needed to understand the causes of these increases in disparities more clearly, as the reduction in accepted reports, but increasing disparities for Māori, could reflect any of these factors [80].

Examining severity arguments, representative studies provide rough approximations of comparisons of disproportionality. One population study showed that the odds of IPV injury—a measure of severity—among Māori women, as noted above, is 2.98 times that of non-Māori women. The disparity in ROCs for family harm in the child protection system, however, is much greater, 8.5 times in 2019 (although the rate of entering the system from any group reduced between 2010 and 2019). This large disparity is further exacerbated at the point of further action required. This difference in disparities between incidence of IPV injury for Māori in the general population and those entering the child protection system via concerns regarding IPV suggests some form of bias is operating, either directly or instrumentally. An alternative explanation is that increasing disparities in the child protection data are an artifice of increased severity of IPV for Māori interacting with an increasing threshold for Oranga Tamariki at intake (i.e., at the report of concern). Whichever it is, it is fair to conclude that increased incidence alone would struggle to account for the size of the disparity at either decision point, meaning instrumental biases are likely at play.

Potential instrumental biases that have some limited evidence are the lack of conceptual clarity around intimate partner violence and its effects on children; and the concept of cumulative harm relied on in guidance that focusses on children but does not explain the causes of IPV or its social context. The construction of IPV as a type of child abuse within a child-focussed system has several effects on the reproduction of inequalities [78]. Firstly, while general definitions of child abuse and neglect now include exposure to IPV as a type of abuse, tensions exist as to its application in practice [76]. As there are differences in the nature and extent of the effects of exposure on both child and adult victims, it is difficult to state with any certainty the effects of exposure, and therefore make best guesses about future harm, or risk of harm, to children [85,86]. Further, the weighing up of risk of harm against the value of both attachment to the non-perpetrating parent, and the culturally variable value attached to maintaining family connections, affects how risks and benefits of intervention are considered in IPV cases [87]. This conceptual terrain affects Indigenous and ethnic minority families differentially. This is because the construction of the nuclear family and individual rights based in Western individualism (for example, ‘best interests’) can swing risk assessments towards intervention when family and extended family relationships valued in Indigenous cultures are viewed as less important by key powerholders and decision makers [88]. Where there are contested and unclear applications of concepts in practice, particularly as they affect thresholds, this provides fertile ground for biases of many kinds to affect decision making as noted above [89].

Secondly, due to the contested nature of the level of harm at which child protection services should become involved, child protection practitioners themselves have diverse attitudes towards whether the child protection service should intervene, leading to codification of responses based on more superficial assessment mechanisms such as numbers of reports (see below). On the other hand, it could be argued that the rising threshold reduces biases, as increased severity makes interventive rationale more clearly connected to serious harm to children around which there is a more general consensus for intervening.

An example of the codification of risk attached to IPV is the concept of ‘cumulative harm’ used in CP guidance that equates recorded incidents (police callouts) with risk; yet this does not apply to those not notified or who do not call police, although this group constitutes the majority of IPV. Cumulative harm, relied on alone, notes the possible impact on the child, but without assisting practitioners to understand the causes of IPV or its social contexts, this can increase interventions that increase inequities. Where explanations are gender free, context free, and not based on theories of IPV, this can inadvertently increase disparities as effective prevention may be overlooked, mothers blamed, and children viewed in isolation. While the guidance has changed recently, until recently the Oranga Tamariki guidance stated that “Family violence is a common factor in the lives of children and young people who are affected by cumulative harm. The presence of violence is highly damaging to the developing child, and a growing body of evidence suggests infants are particularly vulnerable. Alongside the act of physical violence, other forms of maltreatment contribute to cumulative harm including emotional violence (e.g., humiliation, coercion, degradation), indirect impacts on parenting capability (e.g., anxiety and depression undermining a parent’s ability to care for a child), and physical incapacitation as a result of an assault. Practice implication: Parents/caregivers must be assisted to understand the impacts of family violence on their children, on themselves as people and parents and on their relationships with others” [90].

This guidance makes no mention of either the social factors contributing to intimate partner violence, nor theories of intimate partner violence that may provide practitioners with tools to help address its causes and understand patterns of coercive control. The use of the term ‘parents’ and directives to help them both understand ‘the impacts’ on children suggests a power-blind approach that does not recognize the effects of gender, power and coercion on victim-survivors. Nor does it present a nuanced understanding of the effects on survivor parents, who can often maintain positive protective parenting with support, and who themselves have a diverse range of outcomes [91,92].

Attachment to the survivor parent is an important protective factor for children exposed to IPV, and should be maintained if at all possible [92]. Instead, it focusses on effects on children, presents them as absolute, and suggests that both parents need to ‘understand its impact’, rather than stating that the perpetrator’s use of violence is the main issue. While there can be extremely negative effects of multiple exposures on children, it remains that this kind of guidance suggests an approach that does not consider the child within the context of their families, the power dynamics of coercive controlling relationships, the context of the child’s own relationships, or the social inequalities affecting the family [93]. This approach is likely to exacerbate inequities because of a lack of response to the social or community factors that contribute to IPV and are disproportionally experienced by Māori and people living in high-deprivation areas. The focus on equal responsibility of parents will also exacerbate the sense of blame felt by Māori women as victims/survivors of violence, exacerbating disengagement from the system.

Social Worker’s Conceptualisation of Risk in IPV Practice: Cumulative Harm, Context Free

The interaction between risk concepts, recorded prior contacts, and the concept of cumulative harm can be seen in the qualitative study by Keddell [78]. She found that practitioners made efforts to ascertain risks and strengths when making decisions about IPV risk, drawing on cumulative harm and strengths-based practice concepts. They defined risk drawing on their cumulative harm guidance. For example, this practitioner noted that “history alongside-on our database, or the police are saying how many family violence occurrences there have been. So that’s factored in, so there’s ‘cumulative harm’ which combined with children being ‘collateral damage’ in the fighting would increase perceptions of risk”.

Strengths were also constructed in specific ways. This respondent, for example, shows how this is interpreted:

“So yeah with-so on the strengths side though-I guess what we see is that the mother and the children are in the women’s refuge. That you know they’re being assisted to get protective orders, they’ve made a statement to the police. Yeah and I guess they are engaging, when we phone engaging with us and letting us know you know exactly hey yeah nah I’m separating, I’ve had enough. A bit more fiddly when it’s a bit of codependence there, and they’re in the room and not making a statement. So those are kind of the strengths that we’re looking at from the woman’s point, tends to be the woman’s point of view…What steps are they taking to ensure their children are no longer exposed to family violence”

[78] (n.p.).

Responses that equate a lack of compliance with police, not leaving a violent partner or applying for protection orders as not ‘taking steps to ensure children are no longer exposed to violence’ shows the codified ways that women are expected to perform safety or strengths. These are problematic in two ways in relation to inequalities. Firstly, the long history of racism within the police and child protection systems has resulted in a lack of trust and legitimacy within large swathes of the Māori population [94]. Black feminist research has also shown that populations with little trust in public systems to respond fairly, view family and partner relationships as well as wider family collectives as safer than availing oneself of the variable or racist responses of state organisations [95]. In this context, a lack of compliance with police or leaving the relationship should not be viewed as evidence of willingness to put one’s children in harm’s way, but as evidence of the result of a reasonable lack of trust at the community level in public institutions [96]. Less trust by Māori women could be equated with a lack of safety.

The second reason this distorted interpretation of ‘strengths’ and ‘codependence’ may contribute to inequities is that it reflects little cognisance of the dynamics of coercive control often present in intimate partner violence, where silence and not leaving a violent partner may be the result of coercion and intimidation [97]. It may also represent the safest option available at that time [98]. The expectation of women to protect children from men’s violence places responsibility on the wrong adult to amend their parenting behaviour, and can reflect the reluctance of social workers to work directly with men they consider dangerous, which results in the responsibility for his behaviour being transferred to the survivor [99].

These processes may create another mechanism of inequity, pushing up the perceived risk for Māori and others living in high-deprivation areas who may have limited trust in state institutions, have cultural values around the importance of whānau that makes leaving more difficult/less desirable, and may perceive leaving as heightening the known risk of escalating violence. They may also have fewer financial resources to enable leaving, nor the means to support themselves should they choose to leave the relationship. For these reasons, the elision of compliance with authorities as a sign that children are safer holds questionable assumptions and may exacerbate inequities in intervention [14]. In conclusion, the issue of IPV certainly has differences in incidence or ‘real’ risk. However, there are a number of credible interpretive processes that may contribute to instrumental biases that heighten the progression of Māori whānau into the child protection system.

9. Conclusions

The social inequities within child protection systems require urgent action to address the unequal outcomes they exacerbate. Examining the causes of these inequities is therefore important. This article makes tentative efforts towards deepening a conceptual understanding of the role of instrumental biases. It is proposed that instrumental biases exacerbate disparities in several ways, and operate in addition to increased risk and direct biases. Instrumental biases include institutional structures and processes, concepts, and risk proxies that increase contact for some communities with the child protection system, while reducing it for others. The notify-investigate system model that consists of a central statutory service that is ‘reported to’ creates the conditions for poor decision making that may exacerbate biases by providing the worst conditions for fair decisions. Within this broad context, several examples of instrumental biases are identified. These include the demand and supply of services (an institutional structure cause), family-specific surveillance (a risk proxy cause), broadening legislative definitions (a concept cause) and the complex responses to IPV (both a concept and risk proxy cause).

The first is due to the structure, type, and quantity of different services that can shape demand and supply in specific ways. Where there is an over-supply of child protection services, with little supply of culturally acceptable prevention services, this can drive up the rates of disadvantaged groups in system contact. Cognitive heuristics and biases lead to proxies for risk that affect some people more than others, especially when report histories or over surveillance can create base rate and other types of fallacies, or heuristics used in practice lead to ritualised or superficial practice affecting some people groups more than others [100]. Family-specific surveillance, bolstered by a reliance on prior recorded contacts, may increase the perception of risk for those families with a lot of ‘history’, and contribute to less rigour in up-to-date assessments of parenting capacity. This may ratchet families subject to historic and current racism further into the system in a manner unrelated to actual levels of risk, create the decision-making conditions for confirmation bias, and unfairly increase system contact for families from racialized minorities living in poverty.

Broadening legislative definitions of serious harm may also exacerbate biases by effectively widening vague definitions to include ‘well-being’ and lowering thresholds for intervention. This makes it difficult to ascertain objective levels of harm on which to base decisions, nor reserve intervention for the most serious of cases around which there is more community consensus. The interconnections between intimate partner violence and the child protection system are complex. The child protection response may contribute to inequities through the codifying of adult survivors’ behaviours as risky or safe depending on their perceived compliance, with little cognisance of power dynamics or a lack of community trust. On the other hand, while disparities in Aotearoa NZ are increasing, overall reports of concern in all ethnic groups have reduced significantly. The knock-on effect of fewer reports overall shows a heightened threshold for action, so the increasing disparities may be an artefact of the rising threshold at intake as greater consensus around severity reduces direct bias (Keddell & Hyslop [30,58]). However, this cannot account for the size of the disparity, which is marked, suggesting direct and/or instrumental biases. These may reflect concepts used in guidance that do not take into account levels of trust in public institutions, theories of intimate partner violence or the social context of IPV.

Intersecting inequities in the child protection system have multiple causes, including structural risks, direct biases and instrumental biases. The current system requires systemic change to move away from a notify-investigate system to a model that recognizes the social determinants of system contact and addresses them through social protections and community and family-led service provision. This structure would go some way to address both disproportionate risk by providing services that address risk causes, and reduce instrumental and direct biases by reducing the need for decisions made under poor conditions that exacerbate the likelihood of bias. Understanding exactly how particular institutional arrangements and interpretive processes contribute in the form of instrumental biases requires ongoing research to properly identify and address inequitable consequences in differing national contexts.