Anne Milgram: Why smart statistics are the key to fighting crime

220,711 views ・ 2014-01-28

TED


Please double-click on the English subtitles below to play the video.

00:12
In 2007, I became the attorney general
0
12843
2591
00:15
of the state of New Jersey.
1
15434
1725
00:17
Before that, I'd been a criminal prosecutor,
2
17159
2280
00:19
first in the Manhattan district attorney's office,
3
19439
2681
00:22
and then at the United States Department of Justice.
4
22120
2650
00:24
But when I became the attorney general,
5
24770
2201
00:26
two things happened that changed the way I see criminal justice.
6
26971
3895
00:30
The first is that I asked what I thought
7
30866
2030
00:32
were really basic questions.
8
32896
2186
00:35
I wanted to understand who we were arresting,
9
35082
2856
00:37
who we were charging,
10
37938
1664
00:39
and who we were putting in our nation's jails
11
39602
2128
00:41
and prisons.
12
41730
1416
00:43
I also wanted to understand
13
43146
1648
00:44
if we were making decisions
14
44794
1329
00:46
in a way that made us safer.
15
46123
2518
00:48
And I couldn't get this information out.
16
48641
3252
00:51
It turned out that most big criminal justice agencies
17
51893
3357
00:55
like my own
18
55250
1302
00:56
didn't track the things that matter.
19
56552
2382
00:58
So after about a month of being incredibly frustrated,
20
58934
3318
01:02
I walked down into a conference room
21
62252
1971
01:04
that was filled with detectives
22
64223
1890
01:06
and stacks and stacks of case files,
23
66113
2782
01:08
and the detectives were sitting there
24
68895
1176
01:10
with yellow legal pads taking notes.
25
70071
2234
01:12
They were trying to get the information
26
72305
1586
01:13
I was looking for
27
73891
1218
01:15
by going through case by case
28
75109
2045
01:17
for the past five years.
29
77154
1898
01:19
And as you can imagine,
30
79052
1653
01:20
when we finally got the results, they weren't good.
31
80705
2643
01:23
It turned out that we were doing
32
83348
1655
01:25
a lot of low-level drug cases
33
85003
2020
01:27
on the streets just around the corner
34
87023
1475
01:28
from our office in Trenton.
35
88498
2268
01:30
The second thing that happened
36
90766
1467
01:32
is that I spent the day in the Camden, New Jersey police department.
37
92233
3674
01:35
Now, at that time, Camden, New Jersey,
38
95907
1887
01:37
was the most dangerous city in America.
39
97794
2652
01:40
I ran the Camden Police Department because of that.
40
100446
3827
01:44
I spent the day in the police department,
41
104273
2112
01:46
and I was taken into a room with senior police officials,
42
106385
2726
01:49
all of whom were working hard
43
109111
1675
01:50
and trying very hard to reduce crime in Camden.
44
110786
3257
01:54
And what I saw in that room,
45
114043
1826
01:55
as we talked about how to reduce crime,
46
115869
2245
01:58
were a series of officers with a lot of little yellow sticky notes.
47
118114
3859
02:01
And they would take a yellow sticky and they would write something on it
48
121973
2846
02:04
and they would put it up on a board.
49
124823
1799
02:06
And one of them said, "We had a robbery two weeks ago.
50
126622
2171
02:08
We have no suspects."
51
128793
1711
02:10
And another said, "We had a shooting in this neighborhood last week. We have no suspects."
52
130504
5027
02:15
We weren't using data-driven policing.
53
135531
2583
02:18
We were essentially trying to fight crime
54
138114
2042
02:20
with yellow Post-it notes.
55
140156
2527
02:22
Now, both of these things made me realize
56
142683
2135
02:24
fundamentally that we were failing.
57
144818
3251
02:28
We didn't even know who was in our criminal justice system,
58
148069
3123
02:31
we didn't have any data about the things that mattered,
59
151192
3235
02:34
and we didn't share data or use analytics
60
154427
2568
02:36
or tools to help us make better decisions
61
156995
2151
02:39
and to reduce crime.
62
159146
2003
02:41
And for the first time, I started to think
63
161149
2224
02:43
about how we made decisions.
64
163373
1910
02:45
When I was an assistant D.A.,
65
165283
1397
02:46
and when I was a federal prosecutor,
66
166680
1870
02:48
I looked at the cases in front of me,
67
168550
1746
02:50
and I generally made decisions based on my instinct
68
170296
2626
02:52
and my experience.
69
172922
1692
02:54
When I became attorney general,
70
174614
1659
02:56
I could look at the system as a whole,
71
176273
1639
02:57
and what surprised me is that I found
72
177912
1818
02:59
that that was exactly how we were doing it
73
179730
1905
03:01
across the entire system --
74
181635
2303
03:03
in police departments, in prosecutors's offices,
75
183938
2401
03:06
in courts and in jails.
76
186339
2800
03:09
And what I learned very quickly
77
189139
2197
03:11
is that we weren't doing a good job.
78
191336
3633
03:14
So I wanted to do things differently.
79
194969
2016
03:16
I wanted to introduce data and analytics
80
196985
2197
03:19
and rigorous statistical analysis
81
199182
2049
03:21
into our work.
82
201231
1400
03:22
In short, I wanted to moneyball criminal justice.
83
202631
2970
03:25
Now, moneyball, as many of you know,
84
205601
2027
03:27
is what the Oakland A's did,
85
207628
1569
03:29
where they used smart data and statistics
86
209197
1973
03:31
to figure out how to pick players
87
211170
1622
03:32
that would help them win games,
88
212792
1521
03:34
and they went from a system that was based on baseball scouts
89
214313
2980
03:37
who used to go out and watch players
90
217293
1860
03:39
and use their instinct and experience,
91
219153
1637
03:40
the scouts' instincts and experience,
92
220790
1743
03:42
to pick players, from one to use
93
222533
1713
03:44
smart data and rigorous statistical analysis
94
224246
2822
03:47
to figure out how to pick players that would help them win games.
95
227068
3371
03:50
It worked for the Oakland A's,
96
230439
1798
03:52
and it worked in the state of New Jersey.
97
232237
2219
03:54
We took Camden off the top of the list
98
234456
2073
03:56
as the most dangerous city in America.
99
236529
2171
03:58
We reduced murders there by 41 percent,
100
238700
3155
04:01
which actually means 37 lives were saved.
101
241855
2982
04:04
And we reduced all crime in the city by 26 percent.
102
244837
3740
04:08
We also changed the way we did criminal prosecutions.
103
248577
3239
04:11
So we went from doing low-level drug crimes
104
251816
2005
04:13
that were outside our building
105
253821
1642
04:15
to doing cases of statewide importance,
106
255463
2342
04:17
on things like reducing violence with the most violent offenders,
107
257805
3158
04:20
prosecuting street gangs,
108
260963
1858
04:22
gun and drug trafficking, and political corruption.
109
262821
3408
04:26
And all of this matters greatly,
110
266229
2502
04:28
because public safety to me
111
268731
1945
04:30
is the most important function of government.
112
270676
2536
04:33
If we're not safe, we can't be educated,
113
273212
2298
04:35
we can't be healthy,
114
275510
1348
04:36
we can't do any of the other things we want to do in our lives.
115
276858
2945
04:39
And we live in a country today
116
279803
1701
04:41
where we face serious criminal justice problems.
117
281504
3134
04:44
We have 12 million arrests every single year.
118
284638
3661
04:48
The vast majority of those arrests
119
288299
2043
04:50
are for low-level crimes, like misdemeanors,
120
290342
3012
04:53
70 to 80 percent.
121
293354
1734
04:55
Less than five percent of all arrests
122
295088
1991
04:57
are for violent crime.
123
297079
1895
04:58
Yet we spend 75 billion,
124
298974
2055
05:01
that's b for billion,
125
301029
1418
05:02
dollars a year on state and local corrections costs.
126
302447
4127
05:06
Right now, today, we have 2.3 million people
127
306574
2841
05:09
in our jails and prisons.
128
309415
1900
05:11
And we face unbelievable public safety challenges
129
311315
2796
05:14
because we have a situation
130
314111
1939
05:16
in which two thirds of the people in our jails
131
316050
2898
05:18
are there waiting for trial.
132
318948
1754
05:20
They haven't yet been convicted of a crime.
133
320702
2135
05:22
They're just waiting for their day in court.
134
322837
2119
05:24
And 67 percent of people come back.
135
324956
3548
05:28
Our recidivism rate is amongst the highest in the world.
136
328504
3028
05:31
Almost seven in 10 people who are released
137
331532
2103
05:33
from prison will be rearrested
138
333635
1651
05:35
in a constant cycle of crime and incarceration.
139
335286
3955
05:39
So when I started my job at the Arnold Foundation,
140
339241
2582
05:41
I came back to looking at a lot of these questions,
141
341823
2736
05:44
and I came back to thinking about how
142
344559
1654
05:46
we had used data and analytics to transform
143
346213
2383
05:48
the way we did criminal justice in New Jersey.
144
348596
2584
05:51
And when I look at the criminal justice system
145
351180
2144
05:53
in the United States today,
146
353324
1656
05:54
I feel the exact same way that I did
147
354980
1639
05:56
about the state of New Jersey when I started there,
148
356619
2466
05:59
which is that we absolutely have to do better,
149
359085
3228
06:02
and I know that we can do better.
150
362313
1923
06:04
So I decided to focus
151
364236
1705
06:05
on using data and analytics
152
365941
2217
06:08
to help make the most critical decision
153
368158
2361
06:10
in public safety,
154
370519
1606
06:12
and that decision is the determination
155
372125
2021
06:14
of whether, when someone has been arrested,
156
374146
2535
06:16
whether they pose a risk to public safety
157
376681
1915
06:18
and should be detained,
158
378596
1526
06:20
or whether they don't pose a risk to public safety
159
380122
2356
06:22
and should be released.
160
382478
1637
06:24
Everything that happens in criminal cases
161
384115
1919
06:26
comes out of this one decision.
162
386034
1772
06:27
It impacts everything.
163
387806
1496
06:29
It impacts sentencing.
164
389302
1350
06:30
It impacts whether someone gets drug treatment.
165
390652
1901
06:32
It impacts crime and violence.
166
392553
2323
06:34
And when I talk to judges around the United States,
167
394876
1937
06:36
which I do all the time now,
168
396813
1928
06:38
they all say the same thing,
169
398741
1837
06:40
which is that we put dangerous people in jail,
170
400578
3107
06:43
and we let non-dangerous, nonviolent people out.
171
403685
3525
06:47
They mean it and they believe it.
172
407210
2233
06:49
But when you start to look at the data,
173
409443
1733
06:51
which, by the way, the judges don't have,
174
411176
2464
06:53
when we start to look at the data,
175
413640
1612
06:55
what we find time and time again,
176
415252
2418
06:57
is that this isn't the case.
177
417670
1982
06:59
We find low-risk offenders,
178
419652
1681
07:01
which makes up 50 percent of our entire criminal justice population,
179
421333
3714
07:05
we find that they're in jail.
180
425047
2399
07:07
Take Leslie Chew, who was a Texas man
181
427446
2486
07:09
who stole four blankets on a cold winter night.
182
429932
2884
07:12
He was arrested, and he was kept in jail
183
432816
2595
07:15
on 3,500 dollars bail,
184
435411
2053
07:17
an amount that he could not afford to pay.
185
437464
2776
07:20
And he stayed in jail for eight months
186
440240
2588
07:22
until his case came up for trial,
187
442828
2065
07:24
at a cost to taxpayers of more than 9,000 dollars.
188
444893
3905
07:28
And at the other end of the spectrum,
189
448798
1997
07:30
we're doing an equally terrible job.
190
450795
2282
07:33
The people who we find
191
453077
1572
07:34
are the highest-risk offenders,
192
454649
2019
07:36
the people who we think have the highest likelihood
193
456668
2497
07:39
of committing a new crime if they're released,
194
459165
1952
07:41
we see nationally that 50 percent of those people
195
461117
2950
07:44
are being released.
196
464067
1974
07:46
The reason for this is the way we make decisions.
197
466041
3174
07:49
Judges have the best intentions
198
469215
1709
07:50
when they make these decisions about risk,
199
470924
1952
07:52
but they're making them subjectively.
200
472876
2484
07:55
They're like the baseball scouts 20 years ago
201
475360
2146
07:57
who were using their instinct and their experience
202
477506
2131
07:59
to try to decide what risk someone poses.
203
479637
2679
08:02
They're being subjective,
204
482316
1530
08:03
and we know what happens with subjective decision making,
205
483846
3060
08:06
which is that we are often wrong.
206
486906
2743
08:09
What we need in this space
207
489649
1383
08:11
are strong data and analytics.
208
491032
2552
08:13
What I decided to look for
209
493584
1747
08:15
was a strong data and analytic risk assessment tool,
210
495331
2836
08:18
something that would let judges actually understand
211
498167
2764
08:20
with a scientific and objective way
212
500931
2259
08:23
what the risk was that was posed
213
503190
1647
08:24
by someone in front of them.
214
504837
1610
08:26
I looked all over the country,
215
506447
1649
08:28
and I found that between five and 10 percent
216
508096
1942
08:30
of all U.S. jurisdictions
217
510038
1329
08:31
actually use any type of risk assessment tool,
218
511367
2978
08:34
and when I looked at these tools,
219
514345
1625
08:35
I quickly realized why.
220
515970
1860
08:37
They were unbelievably expensive to administer,
221
517830
2690
08:40
they were time-consuming,
222
520520
1528
08:42
they were limited to the local jurisdiction
223
522048
2107
08:44
in which they'd been created.
224
524155
1430
08:45
So basically, they couldn't be scaled
225
525585
1793
08:47
or transferred to other places.
226
527378
2209
08:49
So I went out and built a phenomenal team
227
529587
2237
08:51
of data scientists and researchers
228
531824
2044
08:53
and statisticians
229
533868
1626
08:55
to build a universal risk assessment tool,
230
535494
2845
08:58
so that every single judge in the United States of America
231
538339
2393
09:00
can have an objective, scientific measure of risk.
232
540732
4324
09:05
In the tool that we've built,
233
545056
1658
09:06
what we did was we collected 1.5 million cases
234
546714
2868
09:09
from all around the United States,
235
549582
1698
09:11
from cities, from counties,
236
551280
1644
09:12
from every single state in the country,
237
552924
1511
09:14
the federal districts.
238
554435
1746
09:16
And with those 1.5 million cases,
239
556181
2189
09:18
which is the largest data set on pretrial
240
558370
1940
09:20
in the United States today,
241
560310
1805
09:22
we were able to basically find that there were
242
562115
1865
09:23
900-plus risk factors that we could look at
243
563980
3322
09:27
to try to figure out what mattered most.
244
567302
2866
09:30
And we found that there were nine specific things
245
570168
2081
09:32
that mattered all across the country
246
572249
2235
09:34
and that were the most highly predictive of risk.
247
574484
2977
09:37
And so we built a universal risk assessment tool.
248
577461
3705
09:41
And it looks like this.
249
581166
1445
09:42
As you'll see, we put some information in,
250
582611
2612
09:45
but most of it is incredibly simple,
251
585223
2013
09:47
it's easy to use,
252
587236
1432
09:48
it focuses on things like the defendant's prior convictions,
253
588668
2969
09:51
whether they've been sentenced to incarceration,
254
591637
1979
09:53
whether they've engaged in violence before,
255
593616
2264
09:55
whether they've even failed to come back to court.
256
595880
2393
09:58
And with this tool, we can predict three things.
257
598273
2500
10:00
First, whether or not someone will commit
258
600773
1853
10:02
a new crime if they're released.
259
602626
1565
10:04
Second, for the first time,
260
604191
1664
10:05
and I think this is incredibly important,
261
605855
1861
10:07
we can predict whether someone will commit
262
607716
1923
10:09
an act of violence if they're released.
263
609639
1834
10:11
And that's the single most important thing
264
611473
1887
10:13
that judges say when you talk to them.
265
613360
1807
10:15
And third, we can predict whether someone
266
615167
1828
10:16
will come back to court.
267
616995
1990
10:18
And every single judge in the United States of America can use it,
268
618985
3033
10:22
because it's been created on a universal data set.
269
622018
3812
10:25
What judges see if they run the risk assessment tool
270
625830
2609
10:28
is this -- it's a dashboard.
271
628439
2120
10:30
At the top, you see the New Criminal Activity Score,
272
630559
2848
10:33
six of course being the highest,
273
633407
1929
10:35
and then in the middle you see, "Elevated risk of violence."
274
635336
2403
10:37
What that says is that this person
275
637739
1746
10:39
is someone who has an elevated risk of violence
276
639485
2060
10:41
that the judge should look twice at.
277
641545
1885
10:43
And then, towards the bottom,
278
643430
1336
10:44
you see the Failure to Appear Score,
279
644766
1968
10:46
which again is the likelihood
280
646734
1392
10:48
that someone will come back to court.
281
648126
3013
10:51
Now I want to say something really important.
282
651139
2213
10:53
It's not that I think we should be eliminating
283
653352
2727
10:56
the judge's instinct and experience
284
656079
2244
10:58
from this process.
285
658323
1604
10:59
I don't.
286
659927
1058
11:00
I actually believe the problem that we see
287
660985
2007
11:02
and the reason that we have these incredible system errors,
288
662992
2854
11:05
where we're incarcerating low-level, nonviolent people
289
665846
3087
11:08
and we're releasing high-risk, dangerous people,
290
668933
3172
11:12
is that we don't have an objective measure of risk.
291
672105
2723
11:14
But what I believe should happen
292
674828
1300
11:16
is that we should take that data-driven risk assessment
293
676128
2800
11:18
and combine that with the judge's instinct and experience
294
678928
3041
11:21
to lead us to better decision making.
295
681969
2958
11:24
The tool went statewide in Kentucky on July 1,
296
684927
3303
11:28
and we're about to go up in a number of other U.S. jurisdictions.
297
688230
3351
11:31
Our goal, quite simply, is that every single judge
298
691581
2591
11:34
in the United States will use a data-driven risk tool
299
694172
2192
11:36
within the next five years.
300
696364
2091
11:38
We're now working on risk tools
301
698455
1352
11:39
for prosecutors and for police officers as well,
302
699807
3284
11:43
to try to take a system that runs today
303
703091
2700
11:45
in America the same way it did 50 years ago,
304
705791
2796
11:48
based on instinct and experience,
305
708587
2097
11:50
and make it into one that runs
306
710684
1855
11:52
on data and analytics.
307
712539
2469
11:55
Now, the great news about all this,
308
715008
1921
11:56
and we have a ton of work left to do,
309
716929
1617
11:58
and we have a lot of culture to change,
310
718546
1857
12:00
but the great news about all of it
311
720403
1746
12:02
is that we know it works.
312
722149
1868
12:04
It's why Google is Google,
313
724017
2153
12:06
and it's why all these baseball teams use moneyball
314
726170
2462
12:08
to win games.
315
728632
1781
12:10
The great news for us as well
316
730413
1737
12:12
is that it's the way that we can transform
317
732150
1896
12:14
the American criminal justice system.
318
734046
2321
12:16
It's how we can make our streets safer,
319
736367
2357
12:18
we can reduce our prison costs,
320
738724
2299
12:21
and we can make our system much fairer
321
741023
2067
12:23
and more just.
322
743090
1725
12:24
Some people call it data science.
323
744815
2162
12:26
I call it moneyballing criminal justice.
324
746977
2301
12:29
Thank you.
325
749278
1804
12:31
(Applause)
326
751082
4093
About this website

This site will introduce you to YouTube videos that are useful for learning English. You will see English lessons taught by top-notch teachers from around the world. Double-click on the English subtitles displayed on each video page to play the video from there. The subtitles scroll in sync with the video playback. If you have any comments or requests, please contact us using this contact form.

https://forms.gle/WvT1wiN1qDtmnspy7