Why smart statistics are the key to fighting crime

In 2007, I became the Attorney General

of the State of New Jersey.

Before that, I'd been a criminal prosecutor,

first in the Manhattan District Attorney's office,

and then at the United States Department of Justice.

But when I became the Attorney General,

two things happened

that changed the way I see criminal justice.

The first is that I asked what I thought

were really basic questions.

I wanted to understand who we were arresting,

who we were charging,

and who we were putting in our nation's jails

and prisons.

I also wanted to understand

if we were making decisions

in a way that made us safer.

And I couldn't get this information out.

It turns out that most big criminal justice agencies

like my own

didn't track the things that matter.

So after about a month of being incredibly frustrated,

I walked down into a conference room

that was filled with detectives

and stacks and stacks of case files,

and the detectives were sitting there

with yellow legal pads taking notes.

They were trying to get the information

I was looking for

by going through case by case

for the past five years.

And as you can imagine,

when we finally got the results, they weren't good.

It turned out that we were doing

a lot of low-level drug cases

on the streets just around the corner

from our office in Trenton.

The second thing that happened

is that I spent the day in the Camden,
New Jersey, Police Department.

Now at that time, Camden, New Jersey,

was the most dangerous city in America.

I ran the Camden Police
Department because of that.

I spent the day in the Police Department,

and I was taken into a room
with senior police officials,

all of whom were working hard

and trying very hard to reduce crime in Camden.

And what I saw in that room,

as we talked about how to reduce crime,

were a series of officers with a
lot of little yellow sticky notes.

And they would take a yellow sticky

and they would write something on it

and they would put it up on a board.

And one of them, "We had a robbery two weeks ago.

We have no suspects."

And another said, "We had a shooting in this neighborhood last week. We have no suspects."

We weren't using data-driven policing.

We were essentially trying to fight crime

with yellow post-it notes.

Now both of these things made me realize

fundamentally that we were failing.

We didn't even know who was
in our criminal justice system,

we didn't have any data about
the things that mattered,

and we didn't share data or use analytics

or tools to help us make better decisions

and to reduce crime.

And for the first time, I started to think

about how we made decisions.

When I was an assistant D.A.,

and when I was a federal prosecutor,

I looked at the cases in front of me,

and I generally made decisions based on my instinct

and my experience.

When I became Attorney General,

I could look at this system as a whole,

and what surprised me is that I found

that that was exactly how we were doing it

across the entire system,

in police departments, in prosecutors's offices,

in courts, and in jails.

And what I learned very quickly

is that we weren't doing a good job.

So I wanted to do things differently.

I wanted to introduce data and analytics

and rigorous statistical analysis

into our work.

In short, I wanted to moneyball criminal justice.

Now, moneyball, as many of you know,

is what the Oakland A's did,

where they used smart data and statistics

to figure out how to pick players

that would help them win games,

and they went from a system that was based

on baseball scouts

who used to go out and watch players

and use their instinct and experience,

the scouts' instincts and experience,

to pick players, from one to use

smart data and rigorous statistical analysis

to figure out how to pick players
that would help them win games.

It worked for the Oakland A's,

and it worked in the State of New Jersey.

We took Camden off the top of the list

as the most dangerous city in America.

We reduced murders there by 41 percent,

which actually means 37 lives were saved.

And we reduced all crime in the city by 26 percent.

We also changed the way
we did criminal prosecutions.

So we went from doing low-level drug crimes

that were outside our building

to doing cases of state-wide importance,

on things like reducing violence
with the most violent offenders,

prosecuting street gangs,

gun and drug trafficking, and political corruption.

And all of this matters greatly,

because public safety to me

is the most important function of government.

If we're not safe, we can't be educated,

we can't be healthy,

we can't do any of the other things

we want to do in our lives.

And we live in a country today

where we face serious criminal justice problems.

We have 12 million arrests every single year.

The vast majority of those arrests

are for low-level crimes, like misdemeanors,

70 to 80 percent.

Less than five percent of all arrests

are for violent crime.

Yet we spend 75 billion,

that's b for billion,

dollars a year on state and local corrections costs.

Right now, today, we have 2.3 million people

in our jails and prisons.

And we face unbelievable public safety challenges

because we have a situation

in which two thirds of the people in our jails

are there waiting for trial.

They haven't yet been convicted of a crime.

They're just waiting for their day in court.

And 67 percent of people come back.

Our recidivism rate is amongst the highest

in the world.

Almost seven in 10 people who are released

from prison will be rearrested

in a constant cycle of crime and incarceration.

So when I started my job at the Arnold Foundation,

I came back to looking at a lot of these questions,

and I came back to thinking about how

we had used data and analytics to transform

the way we did criminal justice in New Jersey.

And when I look at the criminal justice system

in the United States today,

I feel the exact same way that I did

about the State of New Jersey when I started there,

which is that we absolutely have to do better,

and I know that we can do better.

So I decided to focus

on using data and analytics

to help make the most critical decision

in public safety,

and that decision is the determination

of whether, when someone has been arrested,

whether they pose a risk to public safety

and should be detained,

or whether they don't pose a risk to public safety

and should be released.

Everything that happens in criminal cases

comes out of this one decision.

It impacts everything.

It impacts sentencing.

It impacts whether someone gets drug treatment.

It impacts crime and violence.

And when I talk to judges around the United States,

which I do all the time now,

they all say the same thing,

which is that we put dangerous people in jail,

and we let non-dangerous, non-violent people out.

They mean it and they believe it.

But when you start to look at the data,

which, by the way, the judges don't have,

when we start to look at the data,

what we find time and time again,

is that this isn't the case.

We find low-risk offenders,

which makes up 50 percent of our
entire criminal justice population,

we find that they're in jail.

Take Leslie Chew, who was a Texas man

who stole four blankets on a cold winter night.

He was arrested, and he was kept in jail

on 3,500 dollars bail,

an amount that he could not afford to pay.

And he stayed in jail for eight months

until his case came up for trial,

at a cost to taxpayers of more than 9,000 dollars.

And at the other end of the spectrum,

we're doing an equally terrible job.

The people who we find

are the highest risk offenders,

the people who we think have the highest likelihood

of committing a new crime if they're released,

we see nationally that 50 percent of those people

are being released.

The reason for this is the way we make decisions.

Judges have the best intentions

when they make these decisions about risk,

but they're making them subjectively.

They're like the baseball scouts 20 years ago

who were using their instinct and their experience

to try to decide what risk someone poses.

They're being subjective,

and we know what happens
with subjective decision-making,

which is that we are often wrong.

What we need in this space

are strong data and analytics.

What I decided to look for

was a strong data and analytic risk assessment tool,

something that would let judges actually understand

with a scientific and objective way

what the risk was that was posed

by someone in front of them.

I looked all over the country,

and I found that between five and 10 percent

of all U.S. jurisdictions

actually use any type of risk assessment tool,

and when I looked at these tools,

I quickly realized why.

They were unbelievably expensive to administer,

they were time-consuming,

they were limited to the local jurisdiction

in which they'd been created.

So basically, they couldn't be scaled

or transferred to other places.

So I went out and build a phenomenal team

of data scientists and researchers

and statisticians

to build a universal risk assessment tool,

so that every single judge in
the United States of America

can have an objective, scientific measure of risk.

In the tool that we've built,

what we did was we collected 1.5 million cases

from all around the United States,

from cities, from counties,

from every single state in the country,

the federal districts.

And with those 1.5 million cases,

which is the largest data set on pretrial

in the United States today,

we were able to basically find that there were

900-plus risk factors that we could look at

to try to figure out what mattered most.

And we found that there were nine specific things

that mattered all across the country

and that were the most highly predictive of risk.

And so we built a universal risk assessment tool.

And it looks like this.

As you'll see, we put some information in,

but most of it is incredibly simple,

it's easy to use,

it focuses on things like the
defendant's prior convictions,

whether they've been sentenced to incarceration,

whether they've engaged in violence before,

whether they've even failed to come back to court.

And with this tool, we can predict three things.

First, whether or not someone will commit

a new crime if they're released.

Second, for the first time,

and I think this is incredibly important,

we can predict whether someone will commit

an act of violence if they're released.

And that's the single most important thing

that judges say when you talk to them.

And third, we can predict whether someone

will come back to court.

And every single judge in the
United States of America can use it,

because it's been created on a universal data set.

What judges see if they run the risk assessment tool

is this: it's a dashboard.

At the top, you see the new criminal activity score,

six of course being the highest,

and then in the middle you
see "elevated risk of violence."

What that says is that this person

is someone who has an elevated risk of violence

that the judge should look twice at.

And then, towards the bottom,

you see the "Failure to Appear" score,

which again is the likelihood

that someone will come back to court.

Now I want to say something really important.

It's not that I think we should be eliminating

the judge's instinct and experience

from this process.

I don't.

I actually believe the problem that we see

and the reason that we have
these incredible system errors,

where we're incarcerating
low-level, nonviolent people

and we're releasing high-risk, dangerous people,

is that we don't have an objective measure of risk.

But what I believe should happen

is that we should take that
data-driven risk assessment

and combine that with the
judge's instinct and experience

to lead us to better decision-making.

The tool went state-wide in Kentucky on July 1st,

and we're about to go up in a
number of other U.S. jurisdictions.

Our goal, quite simply, is that every single judge

in the United States will use a data-driven risk tool

within the next five years.

We're now working on risk tools

for prosecutors and for police officers, as well,

to try to take a system that runs today

in America the same way it did 50 years ago,

based on instinct and experience,

and make it into one that runs

on data and analytics.

Now, the great news about all this,

and we have a ton of work left to do,

and we have a lot of culture to change,

but the great news about all of it

is that we know it works.

It's why Google is Google,

and it's why all these baseball teams use moneyball

to win games.

The great news for us as well

is that it's the way that we can transform

the American criminal justice system.

It's how we can make our streets safer,

we can reduce our prison costs,

and we can make our system much fairer

and more just.

Some people call it data science.

I call it moneyballing criminal justice.

Thank you.

(Applause)