Forecasting Domestic Violence: A Machine Learning Approach to Help Inform Arraignment Decisions

• DOI: http://doi.org/10.1111/jels.12098
• Date: 01 March 2016
• Published date: 01 March 2016

Forecasting Domestic Violence: A Machine

Learning Approach to Help Inform

Arraignment Decisions

Richard A. Berk, Susan B. Sorenson, and Geoffrey Barnes*

Arguably the most important decision at an arraignment is whether to release an offender

until the date of his or her next scheduled court appearance. Under the Bail Reform Act of

1984, threats to public safety can be a key factor in that decision. Implicitly, a forecast of

“future dangerousness” is required. In this article, we consider in particular whether

usefully accurate forecasts of domestic violence can be obtained. We apply machine

learning to data on over 28,000 arraignment cases from a major metropolitan area in which

an offender faces domestic violence charges. One of three possible post-arraignment

outcomes is forecasted within two years: (1) a domestic violence arrest associated with a

physical injury, (2) a domestic violence arrest not associated with a physical injury, and (3)

no arrests for domestic violence. We incorporate asymmetric costs for different kinds of

forecasting errors so that very strong statistical evidence is required before an offender is

forecasted to be a good risk. When an out-of-sample forecast of no post-arraignment

domestic violence arrests within two years is made, it is correct about 90 percent of the

time. Under current practice within the jurisdiction studied, approximately 20 percent of

those released after an arraignment for domestic violence are arrested within two years for

a new domestic violence offense. If magistrates used the methods we have developed and

released only offenders forecasted not to be arrested for domestic violence within two years

after an arraignment, as few as 10 percent might be arrested. The failure rate could be cut

nearly in half. Over a typical 24-month period in the jurisdiction studied, well over 2,000

post-arraignment arrests for domestic violence perhaps could be averted.

I. INTRODUCTION

In this article, we address empirically the potential role of domestic violence forecasts

when, at an arraignment, a judge, commissioner, or magistrate decides whether an

*Address correspondence to Richard A. Berk, Department of Criminology, University of Pennsylvania, 19119;

email: berkr@sas.upenn.edu. Sorenson is Professor in the School of Social Policy and Practice at the University

of Pennsylvania; Barnes is Research Assistant Professor of Criminology in the Department of Criminology at the

University of Pennsylvania.

We appreciate the help of Michael Gallagher, a retired police officer, who currently works with a domestic

violence agency in the jurisdiction studied. He provided important information on the meaning of some of our

variables and on related law enforcement procedures. We also received important assistance on the prosecutorial

context of domestic violence from Marian G. Braccia, Deputy District Attorney in the District Attorney’s Office of

the jurisdiction. Thanks also go to the anonymous reviewers of this article.

94

Journal of Empirical Legal Studies

Volume 13, Issue 1, 94–115, March 2016

offender can be released awaiting a formal hearing on the charges. In the past, forecasts

of “future dangerousness” have sometimes been used at arraignments (Goldkamp &

Gottfredson 1985; VanNostran & Keebler 2009; Arnold Foundation 2013), but there

have been to our knowledge no regularized applications of domestic violence risk assess-

ment procedures in this setting. Our forecasting efforts are part of a larger pretrial

reform initiative in a major metropolitan area.

1

A combination of machine learning and routine electronic information normally

available at arraignment might be able to provide timely and useful domestic violence

forecasts of risk. There are examples of successful forecasting in other criminal justice

settings and for other kinds of crimes (Berk 2012). Moreover, machine learning fore-

casts can be delivered within a real time of several seconds. However, a major question

is whether the information routinely available electronically prior to an arraignment is

sufficiently rich to produce usefully accurate forecasts. We address this question using

data on over 28,000 domestic violence arraignments. The performance of the forecasts

is the empirical focus of this article.

II. BACKGROUND

Although the details vary across jurisdictions, shortly after an arrest or an apprehension

through a summons, there is an arraignment at which the offender receives a written

copy of the charges alleged by the police. These charges typically have been reviewed by

a representative from the district attorney’s office and revised as needed.

2

After the

charges are read, a date for a subsequent hearing is specified during which a judge will

determine whether there is sufficient evidence to require a trial. If the judge rules that

there is sufficient evidence, the offender must enter a plea of guilty or not guilty.

At the arraignment, a court official, variously called a judge, commissioner, or

magistrate, decides whether to detain the offender in jail until the hearing or to release

the offender, sometimes on bond or subject to certain conditions, with the requirement

that the offender return to court on the hearing date.

3

That decision is usually shaped

by two factors specified in the Bail Reform Act of 1984 (Adair 2006): the risk of flight

and the threat to public safety. By law and by court decisions from the Seventh and

Ninth Circuits, both are effectively forecasts that are to be made after carefully consider-

ing a range of factors beyond the immediate charges (Federal Judicial Center

1993:8–10). In practice, the forecasts are commonly made on a magistrate’s subjective

1

It is not fully clear what to call an arrested individual at a preliminary arraignment. At some point, that individ-

ual can become a defendant, but in this jurisdiction, the individual does not have to answer to formal charges

until a subsequent hearing. Therefore, we will use the term “offender” rather than “defendant.” There should be

no confusion in context.

2

The arraignment is sometimes called a “preliminary arraignment,” in part because prosecutors have the option

to unilaterally revise the charges subsequently.

3

In this article, we will use the title “magistrate” rather than judge or commissioner. Regardless of title, the tasks

performed at a preliminary arraignment are essentially the same.

95Forecasting Domestic Violence