Shouting into an echo chamber: confirmation bias and its system conditions in the Hawkes Bay case review

 

“It appears that FACS may regularly omit evidence such as evidence of a parent’s ‘strengths’, the effort a parent has made to address substance abuse issues, or the positive parenting approach of the parent. This has occurred despite there being numerous policy documents that indicate that this approach is not permitted,” (Davis, 2019, p.13).

This was a key finding, not of the Hawkes Bay case, but of the ‘Family is Culture’ review, released in New South Wales last week.

The Australian review, invited by the Minister for FACS, (like Oranga Tamariki) was carried out by Professor Megan Davis from University of New South Wales, into the 1,144 Aboriginal and Torres Strait Islander children in out of home care in the state. The review had access to all case files and was wide-ranging, considering every aspect of the processes leading into care, what happens once children are in there, and how they might get out again. Specifically, it aimed to examine the implementation of Aboriginal placement principles in practice, identify the reasons for the high and increasing rates of Aboriginal children in care, develop strategies to reduce the use of care, and improve family reunification.

One of the many troubling findings of the Australian review was that in some cases there was a deliberate skewing of available information, resulting in a misleading view of families being presented in court and, potentially, unnecessary removal of children. There are some remarkable similarities with the Hawkes Bay case review findings, a review into a 2019 case where an attempt was made to remove a baby at birth. Due to whānau resistance, professional and Iwi advocacy, and media involvement, the baby was not removed, and its older sibling is now also in the process of being returned. The review found that:

• “New information about the mother and whānau that was shared with OT but was not consistent with the historical information appears to have been discounted, overlooked or not pursued”.
• “Unsubstantiated information that should have been more thoroughly considered was instead treated as confirmation of historical concerns”.
•  “A pre-determined view may have already taken place in the early stages of the pregnancy”
• “Insufficient consideration was given to if and how safety planning could have been used with these parents and their whānau, despite evidence it may have been used effectively in this context”
•  “Historical concerns were relied upon to form the basis of decisions for the older child as justification for a set pathway for this child”, and
•  “Strengths were identified but not built upon”.

All these comments point to how information gathering and interpretation can be shaped by a pre-existing view, so that positive aspects are not included in the family narrative, while other information that is negative but perhaps fairly thin (from an untrustworthy source, not verified, out of date or presented in a skewed way), is used to support an initial judgement.

Why does this happen? What are the processes and conditions that allow it when it is so obviously (from the outside) anathema to fair practice and good judgement?

One way of explaining the processes that lead to this is that of confirmation bias or avoidance of ‘decision dissonance’. Confirmation bias is where people prefer to rely on information that aligns with their preconceived ideas and pay less attention to information that does not match. As Boiney et al. state: “Individuals tend to evaluate information consistent with the desired conclusion less critically than contradictory information” (Boiney, et al., 1997, p. 3). Other factors contribute to this generally well-known cognitive bias. How information is searched for contributes to it, not just how it is viewed once found. People tend to look for more information that supports initial decisions, rather than dis-confirming information. A focus on the end decision while a decision is still being made, rather than on the available information, heightens the tendency towards confirmation bias; that is, it amplifies reliance on the information that supports one’s preferred outcome. These selective processes are still subject to ‘reasonableness’ constraints, however, as information search and attention is only biased to the extent that the conclusion can still be considered ‘reasonable’.

In the Hawkes Bay case, it was clear that the decision to remove had already been made early in the timeline. This preferred decision shaped information search practices and how that information was perceived. Accordingly, there was limited inclusion of the family in decision processes and their perspective on the issues at hand was dismissed. There was exclusion of dis-confirming evidence provided by the NGO organisation working with the family, and an over-reliance on negative historical information.

In addition, the decision about the current child can only really be understood in the context of the first decision for the older child (who had been removed at birth the previous year). Given that recent decision, it’s hard to have a full and fair assessment with the possibility of arriving at a different decision so soon – because to do so would threaten the legitimacy of that first decision. The review notes this when it states: “historical concerns were relied upon to form the basis of decisions for the older child as justification for a set pathway for this child”. This is another classic type of decision-making bias – sometimes called a ‘choice-supportive’ bias – where one’s choice is retroactively ascribed positive attributes, while the alternative is viewed negatively.

That said, the decision still had to meet some ‘reasonableness’ constraints, which is reflected in the reliance on two newly added legal principles: keeping siblings together and early permanency and stability. The combination of these principles, together with the assumptions built into the subsequent child amendment (although this family technically did not meet the legal threshold required) allowed the decision to seem ‘reasonable’ despite the skewed pattern of information searching and lack of family inclusion.

But these near-universal cognitive bias processes take over in system conditions that allow them to. What were the conditions that exacerbated this confirmation bias?

This case is a classic example of tension between the ‘removal at the earliest possible opportunity to permanency’ principle (introduced post the 2015 reforms) and the emphasis on requiring participation and inclusion of whanau, hapu and iwi in decision-making, retaining whakapapa, and strengthening families to enable them to care for their children where possible. In ‘best case’ scenarios these principles are balanced in tension, and the progressive nature of practice is well understood (i.e. first try support, then go to removal only as a last resort). But in ‘worst case scenarios’, these competing principles can be selectively used, especially if there are not strong systemic processes in place that guide how the legislation is interpreted.

In this case, cognitive bias and the selective use of legislation was exacerbated by the following conditions:

1. Rapid legislative change, as new amendments can take precedence over existing principles. This can result in availability bias – newer, more ‘available’ provisions are over-emphasised in the complex balancing of principles.

2. Weak systemic accountability that allowed the exclusion of alternative sources of information in the decision-making process. A referral for an FGC was made, but it was never held. There were no home visits to properly assess the family. There were no case consults that included either the NGO workers or the family themselves.

3. A reliance on recorded and outdated information. The most salient information is drawn from the relationship with the whānau, not the records. Supervision practise should draw attention to information imbalance: not just ask ‘justify your decision to me’, but ‘what processes, relationships, and sources of information did you use to reach this decision’?

4. The elision of the stated history of domestic violence with a ‘good enough’ reason to remove, without any depth of exploration as to whether that violence was continuing, what form it took and how the mother and baby could be kept safe together if it was persisting.

5. A creeping rise in the normalising of ex parte orders made applying for a ‘without notice’ order acceptable in this case. This is a type of ‘procedural drift’, where practice increasingly departs from stated rules or guidance. This further reinforced the distortion of information, as it stopped the family being able to provide their version of events at the final opportunity to do so, as all earlier opportunities – hui a whānau, FGC and case consults, had been denied them.

6. The implicit effect of ethnic, class and age biases. It’s hard to see how this process would have gained any traction at all if the family were middle class, older and Pakeha. Reflecting on the effects of social inequalities appears absent.

7. The ‘weaponisation’ of trauma. It is concerning that the traumatic experiences of the parent’s own childhoods are used to construct them as ‘risky’ and to justify the most extreme of interventions, rather than informing support and care. This use of trauma as a concept is especially concerning when it is used to bolster a superficial understanding of risk. The review notes: “Statutory social workers have to be able to consider historical factors in the current setting. They must be able to reflect on positive steps taken to make changes in the lives of parents.…this is particularly important as incremental change can enable parents to safely care for their babies despite their own trauma histories” (p.34).

8. An extreme focus on the child, promoted as ‘child focussed practice’ in guidance since the 2015 reforms, frames removal as a gain, because immediate safety is secured. Long term losses for both parents and child are not viewed as relevant. It also frames the exclusion of parents and whānau from decision-making as legitimate, because ‘the child is the client’.

9. Key resources were not available. Expediency is the evil twin of cognitive bias – it helps it take over in low resource environments, heightening selective application of legislation. There was no FGC coordinator employed at the site for months – crucial for an FGC to be held. There were other key vacancies that squeezed other important roles, like that of whakapapa searcher. Where there are staff shortages, it’s easier to go with existing information about the family, justified by selective legal principles. There is little time to engage with whānau, build the relationship, or seek the views of NGOs. These pragmatic factors contributed to poor assessment depth and inadequate family or hapū/iwi collaboration.

10. A site culture that emphasised “decisive and timely decision making to facilitate stable and permanent care as early as possible, especially for babies” (p. 37). This is a clear statement that the site may have had too much emphasis on these important elements at the expense of family inclusion and support. Our research suggested that sites can differ in their institutional cultural values, specifically where each site is placed on the family preservation – child protection continuum (Keddell & Hyslop, 2019).

11. Lack of understanding about the subsequent child provisions. This appears to have affected both the parents, who were viewed as ‘subsequent parents’ despite not meeting the legislative threshold, as well as other possible whānau caregivers, who were also, mistakenly, classed as ‘subsequent parents’ unable to be caregivers. Despite not meeting these criteria, the ‘subsequent child’ rationale is clearly part of the unfair logic of this case, where the family is automatically assumed to be incapable of care based on earlier behaviour.

12. The broader system structure of a notify-investigate child protectionist system focuses on establishing if a risk threshold has been met (and how to evidence it) rather than on how to work with families to address and reduce risk. Responses within this policy framework emphasise individual caregiver responsibility, rather than embracing what has variously been called a social model, inequalities approach, social paradigm, a public health approach or a child welfare orientation (Featherstone et al., 2018; Bywaters et al., 2015; Keddell et al., 2019; Bilson, 2016; Scott et al., 2016, Gilbert et al., 2011). These latter approaches focus more on social context, preventive social conditions and support services, rather than individual investigation.

All the above conditions meant that the resulting assessment was like shouting into an echo chamber, where no dissenting information was allowed, and the focus was on risk.

How about some solutions?

It’s excellent that without notice applications will now require more scrutiny and approval – this is welcomed as the bar for that type of order sank very low in this case. In other countries this type of case would never have met a without notice threshold. But, as with all variability in practice, you can’t manage it solely via more bureaucratic ‘rule from above’. Where large bureaucracies rely on more and more regulations, procedures and approvals to ensure quality decisions get made, this generates unintended consequences. Meeting the reporting criteria for the decision, producing the report, doing it within the stated time-frame, and getting sign off: these become the new goals rather than maintaining a focus on the higher order goals – family collaboration in decision-making, children’s best interests and balanced and in-depth information gathering.

To counter this ritualising of practice, several other things are needed: an organisational culture where the values and ethics of family partnership practice are evenly embedded; systems that are confirmation bias-aware; and development opportunities for practitioner’s own professional judgement. Creating new layers of management for approval can inadvertently reduce the conditions required to improve practitioner’s own front-line decisions. Tighter control can reduce their confidence and may not address confirmation bias processes in information gathering, nor the need to analyse it effectively. Specific training on how to avoid confirmation bias through considering counterfactuals, and supervision that explicitly addresses the processes that can lead to this are important ways to operationalise fairer decision processes. Unfortunately, the subsequent child legislation encourages confirmation bias, by crystalising assumptions based on historical events into current certainties.

So, what next?

This review is a brave admission, especially given the rigid defensiveness immediately following the video being made public. But without knowing how widespread the practice described in the review is, it’s easy to discount it as an aberrant exception. I want to remind readers of the Australian review I mentioned at the outset – where the reviewer was entirely independent and had access to all relevant case files (and the resources to search them all). Here, we are assured instead that an internal review has taken place of more than a hundred other cases and ‘not one was like this one’. Without that wider review being independent, and made public, it is difficult to take that on face value when a number of whānau and other professionals are stating otherwise. Let’s hope that one of the other upcoming reviews or the Waitangi tribunal claim can offer a robust independent perspective.

Image credits: Peter Dashevici

References

Bilson, A. and K. E. C. Martin (2016). Referrals and Child Protection in England: One in Five Children Referred to Children’s Services and One in Nineteen Investigated before the Age of Five. British Journal of Social Work, 47 (3), 793 – 811.

Boiney, L. G., et al. (1997). Instrumental bias in motivated reasoning: More when more is needed. Organizational Behavior and Human Decision Processes, 72(1): 1-24.

Bywaters, P. (2015). Inequalities in child welfare: towards a new policy, research and action agenda. British Journal of Social Work, 45(1): 6-23.

Featherstone, B., Gupta, A. Morris, K. & White, S.  (2018). Protecting Children: a Social Model. Bristol, Policy Press.

Keddell, E., Davie, G. & Barson, D. (2019). Child protection inequalities in Aotearoa New Zealand: Social gradient and the ‘inverse intervention law’. Children and Youth Services Review. https://doi.org/10.1016/j.childyouth.2019.06.018

Keddell, E. & Hyslop, E. (2019)  Networked decisions: Decision-making thresholds in child protection The British Journal of Social Work.  bcz131, https://doi.org/10.1093/bjsw/bcz131

Scott, D., Lonne, B., & Higgins, D. (2016). Public health models for preventing child maltreatment: Applications from the field of injury prevention. Trauma, Violence, & Abuse 17(4): 408-419.