Another Pandora's box? Some pros and cons of predictive risk modeling

https://doi.org/10.1016/j.childyouth.2014.07.016Get rights and content

Highlights

  • Algorithms using administrative data can identify children at higher risk of abuse.

  • Predictive risk modelling creates opportunities to provide supportive services.

  • Early identification of family problems has benefits and limitations.

Abstract

Early intervention, promoted as being important to the prevention of child maltreatment, is challenged by the difficulty of identifying at risk families before patterns of abuse are established. A way of identifying these families before they reach the radar of statutory systems of child protection is through predictive risk modeling (PRM). Using large datasets PRM tools are able to use algorithms with significant capacity to ascertain and stratify children's risk of experiencing maltreatment in the future. In the process, however, they also identify families who may well benefit from support but are not on a maltreatment trajectory — the so called ‘false positives’ who would not be among those families later identified as mistreating their children. Whilst early identification of families through the use of PRM has the potential to offer opportunities to provide supportive services that could ameliorate future harm to children, it is clear that it also has the potential to mistakenly target and label families as potential child abusers. This article discusses challenges and opportunities associated with the use of PRM in child protection. It briefly discusses the development of PRM in New Zealand, and traverses some of the complex issues as systems attempt to better target limited resources in the context of fiscal restraint.

Introduction

Deciding which children are ‘really’ at risk is an abiding problem for professionals with a child protection mandate. In a field pervaded by uncertainty they must repeatedly make complex decisions about children's safety (Mansell, 2006, Munro, 2011). Such decisions are made in the knowledge that maltreatment can result in long-term emotional problems, serious injury or even death, whilst at the same time knowing that intervention can also be disruptive and harmful — to be undertaken only when strictly necessary. It is clear that in staying focused on prompt, responsible action, professionals perform a tricky balancing act (Shlonsky & Friend, 2007), concerned about overlooking signs that children are unsafe, or overreacting when children actually are safe. Agencies reflect this tension where ‘a preoccupation with risk and its management has engulfed public sector services’ (Macdonald & Macdonald, 2010, p. 1174). In reporting high-profile child maltreatment cases the media have at times exacerbated these issues, evoking outrage and sometimes subjecting practitioners to public vilification (Jagannathan and Camasso, 2011, Kemshall, 2002). In the context of this risk-focused paradigm it is perhaps not surprising that agencies worked to tighten control, making child protection practice a closely scrutinised activity directed and monitored by bureaucratic procedures (Munro, 2005). The introduction of a range of consensus and actuarial risk assessment models has not, however, significantly fortified confidence in child protection decision-making (Schwartz, Kaufman, & Schwartz, 2004).

Early intervention has consistently been promoted as an important way forward when responding to vulnerable children and their families (Australia Research Alliance for Children, Youth (ARACY), 2008, Department for Education, 2003, Dubowitz et al., 2011, Reynolds et al., 2009). A key message of the final report on Munro's review of child protection in England is that ‘preventative services will do more to reduce abuse and neglect than reactive services’ (Munro, 2011, p. 69). The desirability of providing preventive services raises the question of how to ensure that families who need such services actually get them. Recognising that research has shown that the aetiology of child maltreatment is complex, involving not only characteristics and experiences of children, their parents and families but also of features of the community and society in which they live, Dubowitz et al. (2011) undertook a prospective longitudinal study of children from low-income families where the children had no prior involvement with child protective services. The rationale for the study was that, although living on a low income makes child maltreatment more probable, most families in this situation do not abuse or neglect their children. On the understanding that resources seldom permit intervening with everyone (Dubowitz et al., 2011, p. 101) the researchers looked for factors that could be identified by child health professionals or other professionals who routinely see children. Discussing their finding that there are indeed a number of factors that professionals should look out for, including mothers with depression, substance abuse or a low level of education as well as a larger number of siblings in a family and an assessment that a young child has a lower than normal score on a standardized scale of mental development, Dubowitz et al. (2011) suggest that screening families could make it possible to deliver preventive services to those families who most need them. Ideally, they say (Dubowitz et al., 2011, p. 101), ‘such a strategy would complement universal policies and programs, such as those that combat poverty and help support families’.

From a policy perspective, making services available to those who need them most requires an approach that will target families and ensure that they have access to services they need without alienating them by stigmatising them. ‘Proportionate universalism’ or ‘cascading service delivery’ (see for example OECD, 2009) has been proposed as a way of achieving this objective. In this model, services designed to promote the well-being of all children, such as free maternity and infant health care, are used as a way of introducing more intensive or more specialized services for children and families deemed to be at greater risk. For instance, O'Donnell, Scott, and Stanley (2008, p. 329) make a case for a preventive approach based on universal services for all children and families, asserting that universal services must be able ‘to identify vulnerable families early enough to change risky behaviours and avoid pathways to abuse’. Few would argue with this, yet professionals working with high risk populations may find it challenging to identify those families who are particularly ‘vulnerable’ and thus in special need of the extra support (Dubowitz et al., 2011). There is evidence that many children are not identified as at risk even after they have actually suffering harm, especially when the harm is caused by neglect or emotional abuse (Gilbert et al., 2009) nor is there an obvious answer to the question of the best ways in which to accomplish the reorientation of families towards more positive pathways. Reviewing evaluations of a range of preventive programs, Reynolds et al. (2009) found that, whilst some programs appear to have some success in reducing the incidence of certain types of maltreatment, there is relatively weak evidence for the efficacy of preventive programs in preventing the spectrum of harm that results from child maltreatment. For example, a randomized control trial of the New Zealand home visiting program Early Start showed a lower incidence of serious physical assault on children perpetrated by parents taking part in the trial compared with the control group, but no corresponding effect on other types of maltreatment (Fergusson, Grant, Horwood, & Ridder, 2006). Moreover, family level outcomes were in fact better in some respects for the control group in the Early Start trial, including a lower rate of mothers being assaulted than in the group who participated in the program (Fergusson et al., 2006). This is a worrying finding given the concerning emotional impact of witnessing family violence. Recognising that child maltreatment is a ‘wicked problem’ that is not necessarily responsive to linear problem-solving, Devaney and Spratt (2009) nevertheless emphasize the critical importance of identifying and supporting young children likely to have poor outcomes and providing support early. This very early intervention would, it is argued, help agencies to get closer to meeting what O'Donnell et al. (2008, p. 326) describe as a ‘moral obligation’ to ensure that intervention does no harm to children and an even greater obligation to prevent the harm of abuse and neglect occurring in the first place. It is this aim of preventing child abuse from occurring in the first place that has spurred the development of predictive risk modeling (PRM) tools that are able to use algorithms with significant capacity to ascertain and stratify children's risk of experiencing maltreatment in the future. This article will now discuss the development of PRM in New Zealand, and consider some of the issues inherent in its use. It will not describe in any detail the technical aspects of the model (for a full description see Vaithianathan et al., 2012 and Vaithianathan, Maloney, Putnam-Hornstein, & Jiang, 2013). Rather, our purpose is to discuss some of the broader imperatives influencing the development of the PRM initiative, and offer some reflections about the challenges and opportunities of using this technology in the identification of children at risk.

Like many English-speaking jurisdictions, New Zealand has experienced persistently high rates of maltreatment (Duncanson et al., 2009, UNICEF, 2003) and concerning rates of infant death by maltreatment (Child and Youth Mortality Review Committee/Te Rōpū Arotake Auau Mate o te Hunga Tamariki, Taiohi, 2009, Connolly and Doolan, 2007). Like many other countries, New Zealand has been looking at ways in which data can be used to better understand children at risk and as a way to better target limited child protection resources (Mansell, 2006). In 2012 the New Zealand Government commissioned a study, the Vulnerable Children Study, to explore whether it is possible to use administrative data held by government to identify children at risk of maltreatment. The study was undertaken by a cross-university team of researchers based at the University of Auckland's Centre for Applied Research in Economics (Vaithianathan et al., 2012). Under strict confidentiality agreements, the researchers had access to a dataset that linked administrative records from the income maintenance service, and the child protection service, both agencies under the auspices of the Ministry of Social Development (MSD). Both services hold information that is collected on a nationwide basis. A literature review was undertaken, including an extensive international grey literature search, to scope child protection risk assessment generally and previous applications of PRM in particular. This review found only one researcher studying computational intelligence techniques to predict future maltreatment (Schwartz, Jones, Schwartz, & Obradovic, 2008). The researchers then proceeded to develop an algorithm that could be used to predict future harm for children identified within the databases.

The New Zealand study sample comprised children born between January 2003 and June 2006 and whose family received a main benefit (intended to cover basic living costs) for any length of time (termed a ‘benefit spell’) between the child's birth and fifth birthday. The dataset extended to mid-2011, enabling the researchers to retrospectively ‘follow’ children to establish whether or not they had a substantiated finding of maltreatment during their first five years. The data supplied by MSD is routinely collected, relatively easily retrievable and contains much detail, including information about parents' own history of child maltreatment as well as recent information suggesting problems, for example imprisonment, mental health issues or addictions. This detail enabled the researchers to select 132 variables, relating to both the past and the present, for inclusion in the algorithm. It is important to note that the variables are not causal. In devising the algorithm the aim was not to ask ‘What contributes to maltreatment?’ but rather ‘What variables can help us best discriminate between spells that are high risk and spells that are low risk?’ (Vaithianathan et al., 2012). The algorithm was applied at the start of any new ‘benefit spell’; that is, whenever the benefit system recorded alterations in a family's circumstances, such as the arrival or departure of a partner or a shift from one benefit to another. A steep increase in risk rating in succeeding spells signals that problems are escalating. The research question asked whether administrative data could be used to produce a PRM tool capable of correctly assessing the likelihood that a child will have a substantiated maltreatment finding at some future time. The study showed that this was indeed possible. The combined datasets produced a very high ‘capture rate’ in that the families of 57,986 children born between January 2003 and June 2006 received a main benefit. There was a substantiated finding of maltreatment for 11,878 of all children born in this period (5.4%) and the data indicate that the families of 9816 of these children (83%) received an income support benefit before the child turned two. Of these children, 13% were maltreated by age five. Of children whose families did not receive a benefit by the child's second birthday, 1.4% were maltreated by age five. The Vulnerable Children Study model also sorted the children into deciles according to how likely they were to be maltreated during a particular spell and showed that, whilst only 5% of an annual birth cohort will be identified in the top 20% of risk, this group of children accounts for 37% of all children who had substantiated maltreatment by age 5. This was indicative of a very vulnerable group where families were experiencing significant adversity. The study confirmed the disturbing scale of maltreatment in New Zealand.

The study suggests that PRM can help to prospectively identify some children who will later be maltreated, but not all. The top two deciles included 37% of those in the sample who went on to have substantiated findings of maltreatment by age 5. If it is possible to engage families of children in these deciles, and the services are effective, small but appreciable gains could be made in preventing harm. It is important to note that 52% of the children in the top decile (i.e. highest risk) and 71% of those in the second top decile did not go on to have a substantiated finding of maltreatment. The PRM tool can be seen as an indicator of a set of family circumstances that might cause families to experience difficulty and increasing adversity, potentially leading to maltreatment. Families whose children appear in a high PRM decile are likely to have a range of needs that, if met, would improve their circumstances and reduce the likelihood of maltreatment. This has significant implications for the kind of service provision that might be offered to families, who would be in a position to either voluntarily participate in any services offered, or to reject them. Essentially, this means that services offered need to be both useful and appealing. If they are to be used, services must not only meet families' needs in ways that suit them, but be perceived by families as doing so. Otherwise, it is unlikely that families will take up the offer of support (Boag-Munroe & Evangelou, 2012).

Upon release, the Vulnerable Children Study report created a considerable reaction from those working within the area of child and family welfare, human rights, and across the media (NZ Herald, 2012). It seemed that a Pandora's box had been opened. For some PRM represented an infringement of human rights, a form of intrusive surveillance, monitoring families who have shown no sign of maltreating a particular child and creating alerts about some of these families as potentially abusive. For others it offered a genuine opportunity to facilitate early intervention in a powerful counterargument supporting less reactive and more preventive services. These two positions captured well some of the pros and cons of developing PRM tools.

PRM sorts children into deciles according to probability that their family circumstances will become so problematic that their safety is threatened, yet some children in high PRM deciles will never be maltreated. Other children who are experiencing or will experience abuse are not present in either of the two databases used to construct the PRM tool and so are missed by the tool. If PRM is used as part of a child protection strategy but applied only to some children and if these are children appearing on income maintenance and child protection databases then the intensity of surveillance will fall most heavily on a particular cross section of the community. When a database contains disproportionate numbers of people from particular socio-economic or ethnic groups, then families in these groups may be unduly stigmatised and perceived as prone to abusing their children.

From a New Zealand perspective, the possibility of stigma is aggravated by the use of income maintenance data in the prototype PRM tool. A recent survey commissioned by the Human Rights Commission found that ‘74% of people think beneficiaries are facing discrimination’ (Dickison, 2013, p. A3). Using data from the state income support agency and the state child protection agency could exacerbate this, suggesting that benefit receipt is related to personal deficiencies, including a propensity to ill-treat children. This promotes a distorted interpretation of causes of maltreatment. In fact, research has repeatedly shown that maltreatment is associated with an ‘accumulation’ of personal, relational and environmental factors (MacKenzie, Kotch, & Lee, 2011). Families living on benefits commonly experience strain resulting from poverty and may also be struggling with effects of unmet needs resulting from their background and circumstances. Yet most families living in poverty do a good job of parenting (NSPCC, 2008, O'Brien, 2011). Indeed, the Vulnerable Children Study showed that for most of the 57,986 children whose families received income support benefits there was no evidence of maltreatment, whilst 2062 children with a substantiated finding of maltreatment came from families who did not receive benefits. Further, the interaction with agencies required to access benefits means that families who do so are subject to greater surveillance and that data about them accrues. There are, therefore, significant problems with superficially conflating child maltreatment and beneficiary data.

Much maltreatment is hidden and never revealed or discovered and so the true prevalence is unknown (Fallon et al., 2010, Pereda et al., 2011). A key concern is that applying PRM in child protection may create a situation where children who do not appear in a high risk decile are overlooked, whilst compromising the privacy and reputation of children and families categorised as ‘high risk’. Apprehension about the latter issue may be assuaged by assurances that access to PRM scores is restricted and people will see them only for professional purposes. Nevertheless, if PRM scores are shared among a range of professionals across related fields, the sheer number of people thus informed is disturbing. Knowledge of a particular child's high risk score might interfere with how the child and their family are perceived into the future, yet they will not know what information about them was circulated nor how widely (Garrett, 2005). As Langan (2010, p. 96) suggests, ethical and human rights issues arise ‘when service users are unaware that professionals are conducting risk assessments on them and communicating the findings of those assessments to others’. In the context of expectations relating to information sharing across child and family welfare professionals the complexities of maintaining stringent confidentiality become apparent. The counterargument here is that, in certain circumstances, for example when a new partner has a history of serious violence towards children or adults, information should be disclosed for safety reasons, as in a recent trial in the UK (Carter, 2012).

For those ‘false positive’ families identified as representing risk, the ‘risky’ label might nevertheless stick (Pollack, 2010), creating perceptions of them into their future interactions with agencies. Even years later, families who never maltreated their children, and never would have done so, might still carry vestiges of the ‘potential abuser’ label. Writers support the provision of targeted services for ‘high-risk’ groups but advise providers ‘to be sensitive to social inequalities and not to inadvertently widen them’ (Reading et al., 2009, p. 337). Professionals are used to worrying about ‘undermining the privacy of the family, on the one hand, and failing to protect children from abuse and neglect on the other’ (Parton, 2006, p. 63). Yet agencies with knowledge of a child's high PRM score may feel compelled to intervene, fearing the consequences to themselves as well as the child, of not doing so (Webb, 2006). Families may be pressured into accepting help. This is where the complex dynamics of pressure, coercion and discrimination become a real risk for families.

The discriminatory issues relating to using beneficiary and child protection databases present an argument for extending screening to all children, perhaps using data from health or tax agencies. This raises an array of further issues, not the least of which is the potential for notions of risk to completely dominate the field. The potential impact of screening whole populations also creates service response issues. According to Langan (2010, p. 96), in the public health field screening for violence resulted in professionals becoming responsible for ‘managing dangerousness’. A dearth of relevant early intervention services could make statutory child protection agencies responsible by default for responding to high PRM scores, reducing capacity to provide urgent tertiary intervention and remedial services for children who have already been maltreated.

The restriction of PRM to particular groups raises further issues of safety and equity. PRM is not an infallible tool. As noted earlier, it will produce ‘false positives’, attributing high risk to families where children are relatively safe, whilst overlooking others who are unsafe but whose family situation does not feature variables that create a high score. Risk factors are complex and mutable (Schwartz et al., 2008), co-existing and interacting in a state of flux as families' circumstances change and evolve. A PRM score, however frequently repeated, reflects aspects of a family's situation without exploring their meaning for that particular family. Spratt (2008) believes that for economic and social justice reasons, services should be targeted at those most likely to benefit. Some children in unsafe circumstances, however, will not be identified early because their situation does not feature variables included in the model. If PRM is restricted to particular socio-economic groups then some children will not even come within the tool's range and will remain invisible. So, it is a disturbing possibility that professionals and others may come to rely on PRM to pick up risk, dismissing unease about a child as overreaction and drawing back from action that might bring maltreatment to light. Over-reliance on PRM might produce a two-tier professional response where, after a report of concern is made to child protection authorities, children with high PRM scores are investigated more promptly or thoroughly, with less attention paid to children who are scored low or not at all. The latter children may be overlooked entirely, with some languishing in a harmful situation. In a context of fiscal restraint implementation of services targeted at families with a high PRM score may undercut resources for universal services. Vigilance against this possibility is vital, lest ‘invisible’ children and families have even less chance of accessing supportive services.

Typically, at the stage where children are reported to child protection agencies problems are complex and entrenched (Boag-Munroe and Evangelou, 2012, Darlington et al., 2010). Relationship difficulties, disordered attachment, addictions, gambling issues, housing problems and transience are just a few of the collection of troubles correlated with child maltreatment. Devaney (2004) found that a combination of problems is often present in the lives of families whose children are maltreated, the most frequently co-occurring issues being substance misuse and intimate partner violence. The latter is strongly correlated with maltreatment. As Mederos (2004, p. 3) says, ‘the overlap between domestic violence and child abuse is well documented: where one form of family violence exists, there is a likelihood the other does as well’. Many families are caught in the ‘intergenerational cycle’ of deprivation and violence (Reading et al., 2009, p. 337) and experiences of being parented influence children's eventual capacity to parent their own children (Conger, Belsky, & Capaldi, 2009). Children who live in violent households may have less chance of learning positive parenting skills absorbed by more fortunate peers raised in well supported families. Exposure to intimate partner violence is increasingly recognised as a form of maltreatment (Hughes & Chau, 2013). As Teicher (2000, p. 67) explains, ‘Our brains are sculpted by our early experiences. Maltreatment is a chisel that shapes a brain to contend with strife, but at the cost of deep, enduring wounds’.

This well documented research rests at the heart of the argument for early intervention. Pelton (2011) underscores the grinding strain of struggling to meet basic needs. It is not so much poverty per se that undermines parenting capacity but the stress and social deprivation inherent in poverty (Atree, 2005) together with the ongoing effects of parents' personal history and circumstances. Financial hardship is a debilitating aspect of cumulative risk where families become overwhelmed by compounding problems (Waller, 2001) that coalesce in ‘risk chains’ (Rutter, 1999) or ‘cascade effects’ (Masten & Powell, 2003), terms used to describe the phenomenon where one problem leads to another. Arguments supporting early interventions, that provide positive solutions before problems become entrenched, are persuasive. Few would disagree that providing support to families to break cycles of abuse is preferable to trying to fix deeply troubled family situations where children are being abused.

Increasingly technology and the use of administrative datasets provide opportunities that will enable us to identify families and provide a context of early support. An argument supporting the use of PRM to shift child protection towards prevention is to interpret a high PRM score as indicating that a family is, as Langan (2010) puts it, ‘at risk’ rather than ‘a risk’. A high PRM score can be read as indicator of unmet need rather than as an alert that a particular family is a dangerous place for a child. Disconnecting PRM scores from child abuse itself could go some way towards supporting broader early intervention aims. Stigma could be defused through professionals embodying and promoting the notion that attending to needs shows strength. A promising course of action would be to build on existing services designed to meet needs that many families share. It could be argued, nevertheless, that there are moral reasons not to apply PRM in contexts where services are not readily available for families identified as having unmet needs.

As well as identifying families early, PRM will also identify ‘hard-to-reach’ families, those ‘people with significant needs who, for various reasons, do not make use of the support offered to them by statutory agencies’ (Boag-Munroe & Evangelou, 2012, p. 209). In their systematic review of literature on hard-to-reach families, Boag-Munroe and Evangelou (2012) advise providers to consider why families may not want to use their services, citing alienating settings and communication difficulties as likely reasons. For families who have come into contact with child protection agencies another barrier looms — the fear of their children being taken into care. Whilst the children of these families have the poorest long-term outcomes they are nevertheless the hardest to help. These families are likely to avoid services where their situation may be scrutinised. To reach these children professionals will need to consider ‘how best to identify and intervene with families ahead of the traditional intervention points at times of crisis’ (Devaney & Spratt, 2009, p. 640). Agencies that make the attitudinal shift from ‘hard-to-reach’ to ‘how-to- reach’ (Boag-Munroe & Evangelou, 2012) face the challenge of making their services appealing to families who are reluctant, fearful or simply see no value in support available. Introducing a collection of UK studies under the banner ‘Supporting Parents’, Quinton (2004, p. 22) highlights the ‘balance required between the neglect of family problems and intrusion into family life, not to mention differences in individual and cultural ideas of what satisfactory parenting is’.

A good deal of the argument for PRM is based on the notion that families will respond to and welcome services that are provided. Studies undertaken with parents (Duncan et al., 2005, Quinton, 2004) indicate that parents want support that is reliable, trustworthy and builds on strengths. Conversely, they resist or reject services that they experience as demanding, boring, disparaging, or imposed upon them. The low rate of voluntary use of formal support revealed by the Supporting Parents studies was thought to reflect families' desire to maintain autonomy and dread of being ‘taken over’ by authorities. Those most worried about this were the most vulnerable families, who feared that they would be perceived as unable to cope. Families wanted services that ‘took their views and needs seriously, listened to them and were emotionally supportive as well as practically helpful’ (Quinton, 2004, p. 189). This involves developing an integrated, reliable, practical approach (Scott, 2010) including ‘concrete supports’ such as daycare, housing, and financial assistance (Pelton, 2011).

Section snippets

Discussion

Curiously, depending upon how it is approached, PRM has the potential to make systems more, or less risk-dominated. If used to identify individual families it could easily serve to reinforce forensically oriented child protection practice, stigmatising and labeling families as well as overwhelming service responses. In this respect there is real potential for PRM to further reinforce the risk paradigm creating more risk-averse systems that struggle to be responsive to families. Alternatively

Acknowledgments

We are grateful to Associate Professors Rhema Vaithianathan, Liz Beddoe and Tim Dare for the helpful discussions about the content of this article and insightful comments on earlier drafts.

References (55)

  • R. Reading et al.

    Promotion of children's rights and prevention of child maltreatment

    The Lancet

    (2009)
  • D. Schwartz et al.

    Computational intelligence techniques for risk assessment and decision support

    Children and Youth Services Review

    (2004)
  • R. Vaithianathan et al.

    Children in the public benefit system at risk of maltreatment: Identification via predictive modeling

    American Journal of Preventive Medicine

    (2013)
  • P. Atree

    Parenting support in the context of poverty: A meta-synthesis of the qualitative evidence

    Health & Social Care in the Community

    (2005)
  • Australia Research Alliance for Children, Youth (ARACY)

    Inverting the pyramid: Enhancing systems for child protection

  • E. Boag-Munroe et al.

    From hard to reach to how to reach: A systematic review of the literature on hard-to-reach families

    Research Papers in Education

    (2012)
  • C. Carter

    Police launch pilot scheme allowing queries about domestic violence history

    The guardian

    (2012, September 6)
  • Child & Youth Mortality Review Committee/Te Rōpū Arotake Auau Mate o te Hunga Tamariki, Taiohi

    Fifth report to the Minister of Health reporting mortality 2002

    (2009)
  • R. Conger et al.

    The intergenerational transmission of parenting

    Developmental Psychology

    (2009)
  • M. Connolly et al.

    Lives cut short: Child death by maltreatment

    (2007)
  • Department for Education

    Every child matters

    (2003)
  • J. Devaney

    Relating outcomes to objectives

    Child and Family Social Work

    (2004)
  • M. Dickison

    Beneficiaries ‘attacked on all sides’

    The New Zealand Herald

    (2013, February 6)
  • J. Duncan et al.

    Early childhood centres and family resilience

    (2005)
  • M. Duncanson et al.

    Death and serious injury from assault of children aged under 5 years in Aotearoa New Zealand: A review of international literature and recent findings

    (2009)
  • D. Fergusson et al.

    Randomized trial of the early start program of home visitation: Parent and family outcomes

    Pediatrics

    (2006)
  • P. Garrett

    Social work's ‘electronic turn’: Notes on the deployment of information and communication technologies in social work with children and families

    Critical Social Policy

    (2005)
  • Cited by (19)

    • Dissenting social work: Critical theory, resistance and pandemic

      2021, Dissenting Social Work: Critical Theory, Resistance and Pandemic
    View all citing articles on Scopus
    View full text