Family Courts, Risk Assessment, and the Moral Compass

Limitations of Risk Assessment in Relation to Family Courts and Child Protection

The bulk of risk research in psychology and psychiatry has focused on the contribution of mental disorders to the increase in risk. There is much less empirical research into psychological or psychiatric risk factors that directly or indirectly influence the risk of serious harm to a child. Prediction of future risk relies heavily on past events: if we are to project the future based on the past, we need not only good data about the past, but also some confidence that the future will be like the past or some basis for figuring out how it will be different. The need to predict brings us to the question for the present discussion: what are minimally acceptable as good data and how confident do we have to be to talk about future risk?

It is generally believed that the more accurate the data on which risk is based, the more accurate the prediction. The most accurate predictions of future risk are derived from actuarial data (based on statistical analysis of group behavior), as opposed to individual clinical data (i.e., a single clinician assessing and summarizing the risk factors⁴). Usually, concerns about risk start from allegations of behavior, speech, or expressed intentions that are perceived as a threat of violence. Good-quality risk assessment and prediction rely on what are categorized as dynamic factors that change (such as mental state or intoxication) or historical factors, such as criminal record and age. The key point here is that historical factors are fixed and cannot change: The establishment of facts, therefore, has tremendous relevance to risk assessment because of its impact on scoring of actuarial risk measures. For example, a clinician may be asked to perform a risk assessment on an individual who has no established history of violence or arrest. There may be allegations against an evaluee and these may be serious, but there is currently no way for forensic experts to utilize risk assessment tools based on allegations. In fact, any clinician who does so would be going beyond his expertise, misusing the actuarial tool in a potentially negligent way, and acting unethically.

Courts have been traditionally content to accept individual clinical judgment as a basis for risk assessment. Now, however, it is acknowledged that in some circumstances, clinical judgment alone (i.e., an individual assessment of the evidence) is no better than a “flip of the coin” (Ref. 4, p 31). Mulvey and Lidz⁸ argued persuasively that it is quite plausible that unstructured clinical judgments are not made consistently and rationally. Skeem and colleagues⁹ have found that assessments made solely on clinical experience often overestimate risk (i.e., have high false-positive rates).

The complication in these scenarios is that where there are disputed facts, judges are asked to make a finding of fact that determines history for the purposes of future risk assessment. An expert in the English family court may find himself asked to provide an assessment of risk, where the only facts are those determined by a single judge as being 51 percent likely to have occurred and therefore are true for the purposes of determining risk to a child and related welfare concerns.¹⁰ Experts are asked to accept the accuracy of the judge's findings, even though an assessment on the standard of balance of probabilities closely resembles the type of unstructured one-off (i.e., single) clinical assessment that is now deemed to be poor practice in clinical forensic settings. The evidentiary standard of the facts determined by a single judge and the weight and morality of the application of facts determined in risk assessments (on which so much depends for a family) is the central theme of this essay.

Judicial Approaches in England and Wales to Risk, Facts, and Evidence in the Family Court

Actuarial violence risk assessments have established that generally speaking, if there is evidence of harm done in the past, then it must follow, if nothing is done, that there is a risk of harm in the future. By the same argument, however, the possibility that there has been harm in the past does not necessarily establish the risk of harm in the future. The strongest statement that can be made is that what may have happened in the past means that there may be a risk of the same thing happening in the future, although the size of the risk is unknown,⁹

However, courts that deal with child welfare have adopted a much more cautious approach. For example, in Re B, heard by the House of Lords in 2008,¹⁰ the court said that a prediction of future harm could be based on findings of actual fact made on the balance of probabilities. In this case, in a much quoted judgment, the then Baroness Brenda Hale said that in the English legal system, if a judge found it more likely than not that something did take place, then it did take place.¹⁰ The occurrence of harm is enough for a child to be removed from his parents, if the parents are suspected of perpetration, even if no perpetrator can be identified.¹¹

In determining whether harm has actually been done by someone and then finding that someone, the English family courts have long debated whether there should be a sliding scale of certitude based on the likelihood of occurrence of the alleged event. In Re H¹² (a sexual abuse case) Lord Nicholls of Birkenhead said that the standard of proof may vary with the gravity of the misconduct alleged or even the seriousness of the consequences for the person concerned. However, subsequent judgments at appellate levels have established as settled law that there is but one civil standard of proof: balance of probability or more likely than not, roughly thought to be over just 50 percent (or a flip of a coin). This standard is applied when establishing the threshold for making a care order under § 31(9) or the welfare considerations under § 1 of the Children Act 1989. Neither the seriousness of the allegations nor the seriousness of the consequences should make any difference in the standard of proof in determining the facts.

It makes intuitive sense to think that the more serious the allegation, the less likely that the event occurred and, hence, the stronger the evidence needed before the court concludes that the allegation is established on the balance of probability. Ungoed-Thomas expressed this neatly in Re Dellow's Will Trusts: “The more serious the allegation, the more cogent is the evidence required to overcome the unlikelihood of what is alleged and thus to prove it.”¹³ Such an approach is consistent with the criminological data, which show that serious physical violence is a comparatively rare form of criminal behavior.¹⁴

The degree of certitude about facts is therefore different for criminal and civil (family) cases, with the former requiring evidence beyond a reasonable doubt and the latter considering only which side has the preponderance of evidence (i.e., whether the proposition is 51 percent or more likely to be true). The ethics-related argument seems to be that there should be a very high standard of proof before a person is found guilty of a criminal act, but a much lower one for determining a history of (possibly criminal) harm to a child. A single family court judge can, in effect, arrive at a legal opinion that one or both parents have harmed their child, based solely on a reading of the material before him.

From an empirical risk assessment point of view, this is morally unjustifiable. An expert who carried out a risk assessment using this method would rightly be criticized. It might be argued that judges have a public mandate to make these rulings that experts do not have. Judges are aware of this concern, however, as demonstrated (somewhat inconsistently) by Lady Hale in Re B. She quotes Butler-Sloss in another case, Re M and R: The fact that there might have been harm in the past does not establish the risk of harm in the future. Although this is not what the medical view is from psychiatry, the very highest it can be put is that what might possibly have happened in the past means that there may possibly be a risk of the same thing happening in the future. Section 1 (3)(e), however, does not deal with what might possibly have happened or what future risk there may possibly be. It speaks in terms of what has happened or what is at risk of happening. Thus, what the court must do (when the matter is in issue) is to decide whether the evidence establishes the harm or the risk of harm [Ref. 10, para. 50].

She went further: To our minds there can be only one answer to this question, namely the same answer as given by the majority in Re H. The court must reach a conclusion based on facts, not on suspicion or mere doubts… [Ref. 10, para. 50].

Lady Hale is invoking a proper respect for doubt and uncertainty in the finding of facts. Lessons from quantum mechanics suggest that there is a degree of uncertainty in the measurement of individual atoms. Equally, individual humans differ from groups. Statistics for groups work well for car accident insurance, but not for determinations that have massive human consequences. Some human phenomena do not obey the deterministic, Newtonian rules, but resemble chaotic systems where tiny random changes have large, unpredictable effects on other systems.

Family Disputes and Risk

The situation in the family courts is further complicated by the nature of the disputed facts before them. The most common scenario would be one in which a mother accuses an ex-partner of domestic violence or abuse of their children in a custody dispute. Fathers rarely make equivalent allegations of serious violence. Instead, they may claim that a mother is mentally unstable or disordered, has an addiction problem, or is in some way neglectful of the children.

The current adversarial system in England and Wales encourages each parent, based on descriptions of alleged past behavior, to imagine how the other parent will behave in the future. In effect, these are also informal risk assessments based on a parent's fears and concerns. Research on the hidden nature of much domestic and sexual violence toward women has encouraged courts to take seriously allegations of violence made by mothers, even where there is no corroborative evidence. In the United Kingdom, there are pressure and protest groups that have argued that to do so is sexist and therefore unjust.¹⁵ It can be difficult for a man to demonstrate that he presents no risk to others, and current risk assessment tools can only make findings of low risk, not no risk. Women's capacity for antisocial attitudes and behavior is often minimized or overlooked when compared with that of men.¹⁶

Consider the following imaginary case. A couple separate, and their child is left with the husband's family. The mother has a history of mental illness; the father has been the subject of investigation for alleged fraud, but has no convictions for any crime. After several years, the mother returns and seeks sole custody of the child. As part of her case for custody, she alleges that she was the victim of repeated severe violence, both physical and sexual, which the father denies. She also claims that she has received death threats from both her family and that of her ex-husband and consequently is in police protection with a new identity. There is a lengthy fact-finding hearing, in which a single judge finds that all the mother's allegations are fact on the balance of probabilities. An expert is jointly instructed (by all parties, including welfare services) to assess the father and the paternal grandmother, to provide an opinion as to the risk that the father poses to the mother, his child, and women in general.

How should an expert approach this question, morally and legally? In legal terms, the father is now established as a violent rapist, but there is no evidence beyond the materials before the judge, to which the expert does not have full access. Ethically, the expert is at risk if he attempts to make a prediction based on poor-quality data, including the use of actuarial risk prediction instruments, without a proper history. In any event, the use of these instruments is problematic, as the expert is transferring probability data from a group to an individual in the group. For example, an individual score of 22 on the VRAG (Violence Risk Assessment Guide)¹⁷ indicates that a person with the father's characteristics is part of a group with a, say, 35 percent chance of recidivism over a five-year period. What the VRAG does not and cannot tell us is whether this individual will be in the 35 percent who will reoffend or one of the 65 percent who will not. Such statistical uncertainty may work in therapeutic settings where the treating doctor can discuss with the patient this fallacy of prediction based on a group model, and the patient can choose whether to accept the remedial advice. In short, the patient is allowed to take a chance. In forensic assessment of risk, the evaluee is not allowed to take a chance or even express a choice.

Steadman and colleagues¹⁸ in 1994 and Heibrun¹⁹ in 1997 argued that psychiatrists and psychologists have a responsibility to assess the violence risk of their patients or clients, to define the context (when and under which conditions), and to monitor changes in those conditions. The relevant question has now become how we can do justice to this heavy burden we carry and how to do it in a morally defensible way.

One-off clinical risk assessments are thought to be unreliable because of the biases of clinicians and the unreliable weight that they may give to certain factors, as opposed to the more accurate weight given by actuarial data. Judges who establish facts in cases involving disputed violence are effectively making risk assessments and may be equally prone to bias and prejudice. In one case known to me, the judge included a relative's moral compass (or lack thereof) as relevant to the balance of probabilities of whether violence had occurred between a couple.

Then there is the problem of identifying an individual, who may have done nothing, from a pool of perpetrators²⁰ (in Lancashire v. B). Baroness Hale said: Day after day, up and down the country, on issues large and small, judges are making up their minds whom to believe. They are guided by many things, including the inherent probabilities, any contemporaneous documentation or records, any circumstantial evidence tending to support one account rather than the other, and their overall impression of the characters and motivations of the witnesses [Ref. 10, para. 31]. She acknowledges that the task is a difficult one, but then this task is one that the judges are paid to do to the best of their abilities. In the family court, if a judge gets his risk assessment wrong, not only will a child potentially be placed with the wrong caregiver, he will be separated from a really loving one to whom he may have a strong attachment. Disruptions of childhood attachments have negative long-term effects on children's psychosocial development and are just as damaging as any form of abuse. Indeed, Main and Hesse²¹ argued that child abuse is damaging precisely because it is the child's attachment to the abuser that is damaged.

In the hypothetical case described above, any morally defensible forensic risk assessment of the father has to take account, not only of the findings of fact by the judge (made on a mere balance of probability), but also of the more incontrovertible lack of contemporaneous evidence, such as medical records that document violence or police logs or criminal records, or else it remains unsatisfactory. How should the court view the risk? Should it stick by its own findings and any sense of risk that comes from them? Or should the court respect the results obtained by a professional who has carried out a rigorous risk assessment based on all the data there are, including the judge's findings of fact? Is fact determined on balance of probability more probative in risk assessment than absence of evidence that is uncontested? Clinicians often overstate risk for their own safety. Should judges do the same?