Does a simple algorithm help against domestic violence?

Police officers in the German state of Mecklenburg-Western Pomerania predict the likelihood of a future incident of domestic violence using ODARA, a Canadian tool with unproven reliability. It helps cooperation with social workers, but does not work all the time.

For people trying to escape domestic abuse, being dismissed or deprioritized is a matter of life and death. In the last five years, police officers and domestic violence support services in Mecklenburg-Western Pomerania, Germany’s poorest state in regard to the gross regional product per capita, have started to rely on questionnaires as a way to predict the likelihood of a future incident of domestic violence and thus prioritize the highest-risk cases. 

ODARA, or the Ontario Domestic Assault Risk Assessment, is a paper questionnaire. After an incident that comes to their attention, police officers ask the victim 13 questions and add up the “yes” answers.The questions relate to domestic and non-domestic criminal history, children in the relationship and substance abuse. The result, a number between 0 and 13, helps them classify the risk of future violence. 

ODARA was built in 2004 by a group of Canadian researchers. They based it on the information available on 589 men known to the Ontarian police as domestic abusers of female partners or ex-partners.

The questionnaire has been validated in Canada, Switzerland, and Austria. However, a 2016 study on male and female domestic violence offenders (the vast majority of domestic violence perpetrators in Germany are male) who were known to German police found that, of the 13 questions, only two were reliable at predicting the offenders’ chance of reoffending. Moreover, one question - whether the victim is worried about being assaulted by their partner again - was left blank in over 60% of the evaluated questionnaires (In many cases, officers fill out the questionnaire without asking the victims the ODARA questions directly).

In recent years, several automated decision-making systems have been created to predict future incidents of domestic violence. Systems like DyRiAS-Intimpartner and VioGén have been criticized for biases. But in Germany, domestic violence support systems and police are choosing to rely on paper questionnaires like ODARA to predict recidivism. And these have biases of their own.

Risk assessment 

In Germany, police are legally required to make a “risk assessment” when called to a scene of domestic violence. In some states, they use procedures that support services and the Council of Europe criticize as “not standardized”, i.e. officers use individual methods to judge whether a case is urgent, which makes it easier to forget to consider certain common risk factors (e.g. if a perpetrator was convicted for domestic violence in a previous relationship). 

The police in Mecklenburg introduced the ODARA risk scoring system in 2018. Police officers fill out the questionnaire after any incident of domestic abuse. Officers are also using the questionnaire in the German states Baden-Württemberg, Lower Saxony, and Rhineland-Palatinate. In Rhineland-Palatinate and Mecklenburg-Western Pomerania, when officers judge cases as "high risk" via the ODARA questionnaire, they organize regular "interdisciplinary" meetings with lawyers and aid organizations to discuss who can do what to protect the victim from a future offense. Officers use the risk scores to help decide on measures like whether to ban the perpetrator from re-entering a shared flat or house, and the length of that ban. Police also share the risk scores with local domestic violence support services, or “Interventionsstellen”, so that caseworkers can get in touch with victims to offer advice about how they could try and protect themselves from further violence, for example by speaking to neighbors who seem trustworthy and can hear what is going on. 

Meckenburg’s support services do their own risk assessments. In 2017, they tested a more complex algorithm called DyRiAS-Intimpartner. That risk-assessment tool is developed by a German company and runs on a computer. The services decided against it, in part because of its cost. They chose ODARA because that was what the police had decided to use. 

Confusing scores

However, according to Inge Höcker, who helps run the Interventionsstelle in Stralsund, a port city of 60,000 inhabitants in Mecklenburg-Western Pomerania, there are certain high-risk domestic violence situations in heterosexual relationships (ODARA is only focused on violence by a male against his female partner) that ODARA is unlikely to recognize. It focuses heavily on pregnancy, children, and the abuser’s previous offenses, which leaves many blind spots. The ODARA questionnaire won’t consider cases high-risk “where someone who was not previously known to the police suddenly buys a lot of guns” (or other weapons) and “cases where both parties are very young, like 18” and the girl says she’s been choked several times to the point of unconsciousness. Even though, in Höcker’s experience, these are cases where “we need to act immediately.” 

Moreover, even though ODARA is explicitly only for domestic violence by a male against his female partner, police officers are legally required to also fill out the ODARA questionnaire for domestic violence cases between same-sex couples or where an adult child is abusing an elderly parent, for example. These scores are confusing for support services, who are not allowed to receive the police’s filled-out ODARA questionnaires, since changes to the state’s data protection practices in 2020.

Did you like this story?

Every two weeks, our newsletter Automated Society delves into the unreported ways automated systems affect society and the world around you. Subscribe now to receive the next issue in your inbox!