Narrative and fixed-field Data: Are we underestimating the risk of family and domestic violence?

Abstract

Family and domestic violence (FDV) has increasingly been recognised as a major societal issue globally (World Health Organisation, 2021). As research continues to highlight the nature and extent of FDV, growing attention has turned to the impact of FDV on children and young people's safety and wellbeing, highlighting that those exposed to FDV experience a multitude of long-term internalising, externalising and trauma symptoms (Evans et al., 2008; Jenney & Alaggia, 2018). It is estimated that between 133 to 275 million children globally are exposed to at least one incident of FDV each year (Pinheiro, 2006). More recent localised estimates suggest that in the US 17.3 per cent of children had witnessed assault between parents/caregivers in their lifetime (Finkelhor et al., 2013), while across low-income and lower-middle-income countries children's exposure to intimate partner violence was estimated to be 29 per cent (Kieselbach et al., 2022). Considering this, global changes have been enacted to improve child protection policy and legislation and better reflect children and young people exposed to FDV as at risk and in need of protection (Australian Institute of Health & Welfare [AIHW], 2021a; Black et al., 2008).

In many Australian states and territories, exposure to FDV is formally recognised in legislative definitions as grounds for a child in need of protection. However, exposure to FDV is not always represented as its own category of harm in data. For example, Australian population-level statistics report four harm types including emotional abuse, physical abuse, sexual abuse and neglect (AIHW, 2021a). Exposure to FDV is commonly incorporated within the definition of emotional or psychological abuse (AIHW, 2021a; Australian Institute of Family Studies [AIFS], 2019). Australian child protection population-level statistics therefore report on emotional abuse broadly, inclusive of children exposed to FDV (AIHW, 2021a).

Researchers have examined children's exposure specifically to FDV using fixed-field child protection administrative data at the jurisdictional level. These data are recorded as structured, readily extractable pre-set fields (called fixed fields) in an administrative child protection system. For example, Shlonsky et al. (2019) found that 16 per cent of reports to child protection in New South Wales, Victoria, and Western Australian between 2010/11 and 2014/15 related to FDV, with an increase of 25 per cent across that time. Similar rates have been found in prenatal reports to child protection in the Australian Capital Territory, where FDV was the reason for 13.4 per cent of reports (Taplin, 2017).

While fixed-field data provide an indication of the extent of children's exposure to FDV, relying on these high-level operational data is limited for two key reasons. First, operational data are not primarily collected for research purposes. Instead, the information is collected and recorded, typically by social workers, as a part of the day-to-day operation and service delivery of child protection departments. In requesting fixed-field data from data custodians within child protection departments, researchers select the relevant data from an established list of available fields used in client management systems. Due to the availability of information routinely recorded about harm type in fixed-field data, using these data researchers typically can only report on one (primary) or two (primary and secondary) types of harm. However, it is the norm for children known to child protection to experience multiple forms of harm (Moore et al., 2015; Price-Robertson et al., 2013). While relevant child protection authorities can access the nuanced report details to understand the (potential) multiple harm types, researchers using fixed-field data cannot sufficiently capture a complete picture of the child's experiences. Secondly, the recording of a primary and secondary harm type often requires a report to be ‘screened in’ for a child protection response (i.e. those meeting the threshold for abuse and neglect), meaning that ‘screened out’ reports are not examined. By definition, screened-out notifications should not contain abuse or neglect. However, the 2016 South Australian Child Protection Systems Royal Commission assessed a small selection of notifications screened out as notifier-only concerns (i.e. the notification is insufficient or vague, the notifier lacks credibility or the notification does not meet the definition of abuse or neglect), including some that contained concerns relating to children's exposure to FDV. The Child Protection Systems Royal Commission (2016) found that only two of the 20 examined reports would not have required a response from the department. Hence, FDV may be significantly underestimated in studies that use fixed-field information from screened in reports alone.

In recognition of these limitations, research has started to examine the narrative information recorded in child protection reports, further highlighting the potential underestimation of risk. For example, in a recent study in Australia by Meiksans et al. (2021), a much higher level of intimate partner violence (70 per cent) was identified when examining prenatal child protection report narratives compared to that found in previous research using fixed-field data (13.4 per cent; Taplin, 2017). While such a considerable discrepancy between rates of similarly sized Australian samples points to the restrictive nature of fixed-field data, these samples are not directly comparable because they are from different Australian jurisdictions. However, considering the increased reporting of FDV (Shlonsky et al., 2019) this potential underestimation of child exposure to FDV warrants further investigation.

Using fixed-field data, child exposure to FDV was identified in 8.9 per cent (n = 44) of all 493 reports recorded by the child protection department for the sample within the six-month study period. This included 35 reports which had a screening ground of’ ‘significant risk of emotional abuse – domestic violence’ and nine reports with screening ground relating to ‘significant risk of physical abuse – domestic violence’.

While not a focus of this study, to allow comparison with previous research and population-level statistics relating to FDV exposure, emotional abuse as a primary harm type was examined. Seventy of the 493 reports (14.2 per cent) were reported to have a primary harm type of emotional abuse using fixed-field data.

This short report examined the extent of children's exposure to FDV in child protection reports using both fixed-field and narrative data. Results from examination of the narrative data revealed that children's reported current exposure to FDV could be two to three times higher than fixed-field data suggest. In the current study, analyses also showed that over half of all reports to child protection did not have a primary harm type recorded (i.e. the reports were ‘screened out’). A quarter of these included details of child exposure to FDV in the narrative. The implications for these findings are substantial. Underestimating risk is likely to lead to a significant underestimation in the need for services and interventions specifically relating to FDV for families involved in child protection, for example, parenting post-violence and child trauma responses incorporating exposure to violence. Based on the results of the current study, efforts need to be focused on ensuring adequate services are available to support children and families known to child protection who have experienced FDV.

While this study looked specifically at child exposure to FDV, the results may have implications for other types of harm commonly seen in child protection cases, which may be underestimated using fixed-field administrative data. This is an important consideration given that reports of fixed-field data, such as population-level child protection reporting (AIHW, 2021a), are limited to one primary, and sometimes a primary and secondary harm type, despite increasing evidence that experiences and types of abuse or neglect are often multiple, interrelated and co-occur (Moore et al., 2015; Price-Robertson et al., 2013). When exploring harm type in child protection cases utilising fixed field data, consideration should be given to the limitation that it may not capture all types of harm and rather primary harm type typically includes the most serious type of abuse or neglect identified in the case. Future studies should look to investigate data from narrative child protection cases to determine the extent of other factors such as these, particularly where multiple abuse types are present for the child and family.

Narrative data within child protection reports are a rich source of information that in the current study has generated important insights about the extent of child exposure to FDV. Despite this, the methods required (e.g. manual review and coding) take considerable resources, meaning feasibility at a population level and timeliness of contemporary results are limited. Consideration should therefore be given to the use of innovative methods to allow narrative reports to be efficiently examined. Recent advances in data science (e.g. text mining), can automate the processes of reading and coding narrative data, bringing coding to scale and assisting with fast population-level insights to inform research, and assist with decision making for policy and practice. For example, feasibility studies in the US have used text mining and machine learning to identify domestic violence and substance-related concerns in narrative child welfare reports (Perron et al., 2019; Victor et al., 2021). Alternatively, if fixed-field data are to be reliably used in research to inform risk within child protection cases, then recording practices will need to be reformed to include all harm types reported, not just the primary and secondary harm types.

Relying on fixed-field data underestimates the extent of risk and harm to which children are exposed. Fixed-field child protection data are important for efficiently understanding children's contact with the child protection system at a population-level. This paper however demonstrates the limitation of such an approach when it comes to understanding the concerns that are being reported to child protection. Understanding the extent of exposure to risk and harm has implications for demand modelling and service delivery for children and families. Narrative data contained in electronic child protection systems include a wealth of information; utilising innovative approaches to efficiently examine these data is important to understand these concerns.

We have no conflicts of interests to declare.

This study was approved by the Aboriginal Health Council of South Australia, Aboriginal Health Research Ethics Committee (Protocol #04–17-718) and the University of South Australia Human Research Ethics Committee (ID: 0000036590).