Tip:
Highlight text to annotate it
X
>>
All right, on behalf of the authors at Google and the Tech Talks at Google Teams here at
Google Santa Monica, welcome. I'm Christofay and I'm very pleased to welcome Greg Ridgeway
of the RAND Corporation. Ridgeway is the Director of the RAND Safety and Justice Program and
Director of the RAND Center on Quality Policing. He studies the legal firearms markets and
violence prevention initiatives and assist police and communities with law enforcement
and race relations issues. He has worked with a number of major police departments on police
community relations issues including Oakland, New York, Los Angeles and others. Ridgeway
is deeply involved in the research and analysis of contemporary police practices and has received
several commendations and honors for his innovative use of statistics. The title of this talk
today is Racial Profiling Analysis in a Post-Beer Summit World. Please welcome, Greg Ridgeway.
>> RIDGEWAY: Thank you. It's a pleasure to be here today. I'm glad to have such an interesting
audience to hear probably a topic that you don't get to talk much about around here.
So, today, I'm going to be talking about racial profiling analysis and it's something that
I've worked on a lot, but it's come up sort of as part of a line of work we've been doing
for police departments around the country. Police departments don't get a lot of analytical
attention. And many times it's not sort of part of their standard business traditionally.
But now they come across this topic of racial profiling analysis and others are doing their
analysis for them, not always doing it well, so we're going to try to dive into some of
these issues. So racial profiling has been a concern going back several years, certainly
stronger in the minority communities in the U.S. It really came to sort of public prominence
in the 1990s when there is a series of what is known as the I-95 turnpikes studies. The
first one was on the New Jersey turnpike. There was a lawsuit. The Federal Department
of Justice brought a lawsuit against the New Jersey State Police for racial profiling on
the turnpike. And so researchers did something sort of interesting. They collected data on
who was being stopped and the race distribution of who was being stopped in the New Jersey
turnpike. And then, they sent a car going down the New Jersey turnpike going around
60 miles an hour and a passenger in the car would look at the cars that were passing them
and record, "Okay, that looked like White, Black, Latino, Black, Black, Latino, White,
White, White" and then tabulated to see whether the race distribution of who is passing them
in the car was matched at the race distribution of who was being stopped. And they--and they
didn't match. And there were some problems with this study. For example, I mean, if any
of you are from New Jersey, who goes on the turnpike at 60 miles an hour? So no one--no
one goes 60 miles an hour in the turnpike. So everyone was flying past them. And later
studies looked at who was going 70 miles an hour, 80 miles an hour and the results changed
a little bit. The second study was on the Maryland--I-95 through Maryland and that had
similar results. So as a result of these court cases, public concern, the--there's been a
lot of legislation happening. About 14 states have passed legislation banning the practice
of racial profiling, requiring police departments to collect data and that has led to a lot
of different practices going on around the country. Most recently, not so long ago but
two years ago, there was the high profile arrest of Henry Louis Gates and that resulted
in the infamous Beer Summit and that renewed a lot of interest. And, of course, what was
on everybody's mind was what beers will be consumed at the Beer Summit. And I know that
was going to be peak on your interest, so I've listed what actually was consumed at
that Beer Summit. Now, as much as racial profiling was involved in this discussion, it was also
concerned about the status of American beer brewing because many of the selected beers
are now foreign brewed. But that was sort of a sideshow to the main event. So, racial
profiling. There's--it's a question that demands some sort of analysis and there is, in fact,
lots of analysis going on out there. Hundreds of studies perhaps every year are being produced
trying to address the question about whether this department or that department are targeting
minorities in their community. Most of these studies are very weak and they go either way;
either in favor of the hypothesis of racial profiling or against it and there--and there's
several problems. Here are the two of my favorite ones. In Texas, there was a study that concluded
that 75% of agencies stopped more Black and Latino drivers than White drivers. Now, that's
the sort of thing that makes headlines, but let's pause for a minute and think about this.
Well, in Texas, whites are actually the minority now. So, I don't know whether it makes--whether
100% of agencies should stop more Black and Latino just because of the population. Now,
the Black, Latino and Whites are not uniformally mixed throughout Texas. But if, in fact, they
were uniformally mixed, I would expect 100% of departments to pop--stop more Black and
Latino drivers. But there isn't--it isn't sort of uniformally mixed. So I don't know,
in the end, whether that 75% is reasonable or not. There's not a good benchmark to compare
that to. Now, on the other hand, there are some studies that sort of hastily defend the
police department. For example, in Sacramento, the percentage of Black drivers stopped matched
the percentage of Blacks among crime suspect descriptions. And they said, "Okay. Since
that's okay, the department is okay." Now that's not satisfactory either because we
don't stop people for traffic--we stop for traffic offense, rolling through stop signs,
speeding, running a red light. We don't--the cops aren't stopping cars because that person's
an armed robbery suspect or was wanted in a mugging. Now, that happens sometimes, but
there's no reason that those two should match. So it does go both ways. So now, let's get
to some methods that actually are transparent, that are--that really do address the key questions
of interests, and I'm going to focus on three questions; race bias in the decision to make
a stop, internal benchmarking systems that assess whether individual officers are targeting
minorities in their community and lastly looking at post-stop outcomes, search rates, how long
the stops take and so on. So let's start off with a decision to make a stop. So, what makes
this so hard? It seems like a very simple question. We should be able to construct two
pie charts; who's being stopped, who's at risk of being stopped and compare them. Difference
equals ratio profiling. So let's try--make an attempt at this. This is a data from the
City of Oakland 2003, my first study in the area--in this area. In Oakland, they--from
police statistics, they found that 56% of the individual stopped were Black drivers.
Okay. It doesn't--not necessarily we know whether that's too big or too small. We did
some sort of benchmark to compare that to and that's where the big wrinkle is. What
do we compare it to? So the easiest thing is what everybody jumps to is the census and
the two don't match. So 35% of Oakland residents are Black, and so 56% divided by 35% is 1.6
and that's the number that sort of gets touted in the press and in discussions and claims
of a racial profiling. Now there's lots of other reasons why it could be the same--there
could be differences. It could, in fact, be race biased. It could be that there are officers
that are targeting minorities in the community and that might explain some of this. But there
are differences in driving behavior, car ownership, how much--how much time they spend on the
road and the care with which they drive. For example, Hispanics are often are the ones
most likely to wear their seatbelt. If the police are out in force doing seatbelt violations,
they should--actually shouldn't stop many Hispanics in their community. There are differences
across race groups. And lastly, the big one in Oakland is that there are differences in
exposure to the police. That is--in some areas, the flatlands in Oakland, there can be 10
times as many police officers as in the Oakland hills. So if you drive one block through the
flatlands going 70 miles an hour, you're going to be stopped right away. If you drive 70
miles across the--70 miles per hour across the Oakland hills, you probably have a good
of making that without being caught by a cop. You might run into something--some other obstacle,
but not a police officer. So all these things mixed together formulate that difference of
1.6 and we can't attribute that directly to racial profiling. So--but all is not lost.
We have some other methods that we can use. So this is the approach that we took. We noted
the ability to discriminate requires officers to identify the race of the drivers in advance.
That is, think about the racially biased officer. In order to practice racial profiling, they've
got to be able to see the race of the driver in advance and say, "Aha. That's the race
group that I wanted to target. I'm going to go pursue a stop of this vehicle." Okay. Second,
the ability to identify the race in advance of a stop decreases as it gets dark. And this
is something you can test out on your way home this evening or depending on when you
get off from work, that when it's light out, it's reasonably easy to see the race of the
drivers in other cars, not always, but it's easier. And at night, it becomes more difficult.
Now, it's not impossible, but it's more difficult and that's all we need. So, more details of
this are in the paper that I wrote in the Journal of the American Statistical Association.
But using these two features, I'll explain simply what we did. Let's look in Oakland.
Fifty percent of the drivers stopped in daylight were Black, okay? Compare that to the percentage
of Black drivers that were stopped at night, 54%. And so this is actually runs counter
to the racial profiling hypothesis that is if it's dark outside, the racially biased
officer should have more difficulty identifying the race of the drivers in advance. And we
should see that number go down, but instead we see it go up. Now, this naïve analysis
is far from perfect. We're looking at people driving during the day compared to the people
driving at night. People who drive during the day are different from those who drive
at night. The police do different things during the day than they do at night. So, all of
these confound the analysis. So we got--we can't just stop at this simple analysis. We
got to take it one step further. All right. I'm going to show you a graph. On the--on
the horizontal axis, I've got clock time going from 5 p.m. to 9 p.m. at night. On the vertical
axis, I have hours since sunset. And so, we put some dots up there that represents stops.
Each of these is a stop involving the driver. Now you can see, some--like, let's take a
look at this stop right here. This is a stop of a White driver that occurred at around
6:30 in the evening when 6:30 was about two hours before sunset. It's light outside, easier
to see. Now, sometimes, 6:30 occurs after dark, sometimes during the year it's dark
at 6:30. And these stops occurred at 6:30 when it's an hour after sunset and it's dark.
So, we've got this great natural experiment going on that sometimes at 6:30, it's light
outside and sometimes it's dark outside. It gets even better. There's this gap right here
where you see no stops going on. And this is my--this is often puzzling why no stops
occur in this band here and it's because twice a year we suddenly change the clocks from--to
and from Daylight Savings Time. So on one Monday, it's light out at 6:30 and the cops
can see reasonably well into the other vehicles. And the following Monday, it's dark outside.
Now, we don't think the mix of who's driving on the road suddenly switch from one Monday
to the next Monday. We don't think suddenly there's going to be a lot more Black drivers,
more White drivers. The police don't change their patrol pattern just because it's--it's
still at 6:30 not--they don't change their patrol patterns because of Daylight Savings
Time. So we've got this great natural experiment. The only thing different between that one
Monday and the following Monday is that it's light out one Monday and it's dark out the
following Monday, okay? So, what do we do? Slice up this plot into little slivers so
we look just stops right around 6:30. And we see that 53% of the--those stopped at 6:30
once it's light outside were Black and 54% were stopped when it's dark outside, so that's
just for 6:30. We can do this--repeat this analysis across all the different time slices.
We can do some statistical modeling to smooth that out and what we'll find is across the--this
time, we'll find equal rates of Black and White being stopped. So that's the analysis
we did in Oakland. This is the analysis that actually sat around the table with lots of
people from the community, from the police, from the police union, from the ACLU and the
Citizen Police Review Board and all these communities presented this analysis and everyone
sees that, you know, there is not a difference--that the ability to see the race of the driver
in advance is not influencing who's getting stopped. We repeated this analysis in Cincinnati
as well and came up with similar results. So that's looking at bias in the decision
to make a stop. But that's not satisfactory either. In fact, departments and even the
public don't necessarily think that it's a department-wide pattern. They don't think
that the entire department is biased, but I think on both sides there is an agreement
that there are probably a few bad apples in just about every organization of some size.
Okay. And that gets to the second part. If they're--if you're not having problems at
a department-wide level, how can you come up with practices and analysis that will identify
those bad apples before they show up on the six o'clock news? Okay. So this is some work
we did for the New York City Police Department. Here's an officer. This officer made 392 stops
in 2006 of pedestrians, okay? Eighty-six percent of those pedestrians that this officer stopped
were Black. Sounds like a lot or maybe it's not, depends on what part of New York this
officer--what part of New York this officer works. Let's go a little bit more detail on
this officer and what--where--how this officer works. We know that 3% of this officer's stops
occurred in January. We know that 13% occurred on Mondays. We know that 23% occurred between
8 and 10 p.m. We know that they were mostly in Brooklyn North in Precincts B and C, that
is--most of the stops occurred outside. Most of the time this officer made stops in uniform
and so on. We know a lot about this officer. In fact, this is just a partial list. So now
what we can do is look for stops made by other officers that occurred in the similar time
place and context. So, I went through the database of other stops that were made and
I found about 37 stops--100 stops made by other officers that matched the officer in
question's stops. The stop occurred at the same time, same place, same context, okay?
So these officers should have been exposed to the same sort of suspicious characters,
the same sort of infractions and offenses, the same rates of jaywalking. They should
all be seeing the same thing. They're doing the same job. In fact, it's even better, we
have the X-Y coordinates of exactly where these stops are taking place. And this--here's
the officer in question. He looks like he's on a foot patrol, sort of walking up and down
these three blocks. And here is a contour plot of where the benchmark, his peers that
we're using to compare. They're walking up and down that same block too. Okay. So, now
the big question, how--what's the percentage of Black pedestrians that this officer's peers
are stopping? And we find that 55% of those stops involved Black pedestrians. That's a
big difference. Whether statistical or otherwise, that's a--that's a big difference. And now
we've identified a particular problem. And we repeat this analysis for the several thousand
officers that are regularly involved in stopping pedestrians in New York and identified five
officers that are substantially stopping more Black pedestrians than their peers. Then same
for Latino pedestrians found another about eight or so officers that were stopping a
lot more Latino pedestrians than their peers. Now, we've developed a system that's been
deployed in Cincinnati, and I'll talk a bit more about Cincinnati in a second. It's sort
of a web based system that is constantly monitoring officers' activities and they run quarterly
reports. And every quarter, they flag about, say, four or five, six officers and this has
turned out to be a very useful program for them. Now, these six officers that they flag,
the chief has told me, well, let's say they've--they end up falling into three categories. Maybe
one category they knew about. "Yes, there has been a problem with these two officers
that were flagged. We got complaints. We've had some other administrative problems with
them. I've assigned them to a senior field training officer. They've got six months to
shape up or shape out." That's--and that's exactly the sort of thing that we want to
highlight. Another too, they'll have a good explanation, "That officer was on a special
gang detail. We were targeting a specific gang for nine months last year. That's a variable
that is not in this benchmark calculation that we did." Fine. I'm glad they were able
to sort of sort out that question and get that addressed. And the other two officers
they didn't know about. No other signs of problems. No other signs of anything going
on. Let's, you know, mark that. Have the supervisor discuss with that officer and see how things
go over the next--over the next quarter or the next six months. And this is exactly the
kind of discussion that we want police executives to have about managing their police force.
It's not a--necessarily an indictment of the officers that are flagged here, but it's a
step for flagging, among the thousands of officers doing work, can you identify some
of those that require more follow-up? And there's more details about all of the gory
statistical methods behind it, again, in another Journal of the American Statistical Association
paper that came out in 2009. And for the statisticians among you, the methodology, I think, is quite
clever. You might find it useful in your own work. It blends sort of three statistical
methods that are in--being developed, mono-statistical methods that are--the target of a lot of research
that includes the propensity score estimation, doubly robust estimation and false discovery
rates. And this sort of it combines these three methods to identify potential problems
in the police department. The last thing--so we've talked about race bias in the decision
to make a stop, the department-wide, we talked about potential problems with individual officers,
now let's talk about a third step about looking at activity after the stop takes place. There's
a lot of different ways that this can be done. One, we can actually audit police interaction
with the public, and we've done this. We've collected several hundred videotapes from
those in-car cameras that you see on the cop show and we watched them and we had--and we
coded them. We looked for uses of terms of respect on the part of the officer, on the
part of the driver. Did the officer have his hand on his gun? Did he walk with his back?
Lots of things you can code from these. And there's an interesting things that can come
out of that. One interesting thing is we found that Black officers and White officers in
the particular community where we're studying are different. They ask different questions.
In fact, the Black officers tended not to ask many questions. Maybe, you know, a man
who rode through a stop sign, here's your ticket, off you go. Very simple interaction.
The White officers tended to be much more proactive. They saw the stop, it's an opportunity
to do an investigation. Where are you going? Do you have any contraband in the car? Who--ask
IDs from the others in--other passengers in the car. These are two different but valid
approaches to policing. Now if you're a young Black man in this community and you have a
very different experience with the Black officer and a White officer, you're not going to just
attribute that to different frameworks of policing. I think it's reasonable to expect
that that young Black man to attribute that to race bias. So, there are some things that
a department could do to reduce the appearance the--and the perception of a racial--race
bias and their policing. So that's something that we got out of looking at these police
videos. It's often common practice to look at hit rates. After the stop takes place,
do you find--do you do a search and are those searches productive in the sense that you
found drugs, weapons or something like that? That's an interesting avenue to take. So what
we're going to look at is comparing Black and White drivers in similar situations and
look at stop outcomes. And more details are on the paper in the Journal of Quantitative
Criminology if you want to look up the details on this. So let's take a look at what happened
in Cincinnati. Now Cincinnati, you might remember it was about 10 years ago now, was the site
of riots following the shooting of Timothy Thomas. And that led to about a five, six-year
federal oversight of the Cincinnati Police Department. Now, that department is quite
a changed department from where it was 10 years ago. It's winning awards. It's sort
of a model department in any ways. At the time we're doing this analysis, they were
in the middle of a court settlement with a class action lawsuit on civil rights violations.
So let's take a look at what was happening in Cincinnati at that time. If we look at
Black drivers, in one year, about 27,000 Black drivers were stopped in Cincinnati. Fifty-five
percent of them had stops last in less than 10 minutes. And this is about as long as a
stop should take if everything sort of simply you ran a stop light and you run your license
and every--you got a ticket, off you go. Should stop--those stops should last less than 10
minutes. So Black drivers, about 55% of the time, they had one of these stops, these short
stops. On the other hand, non-Black drivers--and I use the term non-Black drivers--in Cincinnati,
that's essentially White drivers. But 65% of them had stops lasting less than 10 minutes.
So the White drivers in Cincinnati were experiencing much shorter stops more frequently than the
Black drivers. So why is this? A big question. Now, this difference is the kind of thing
that makes the headlines. Big difference, big disparity between our Black drivers and
White drivers are treated in our community. But there is more to the story. Let's have
a look at that. Just like with the officers, we know a lot of detail about these stops.
For example, take a look at the driver's licenses. Twenty-two percent of Black drivers stopped
in Cincinnati did not have a valid driver's license. Now, I think everyone can agree that
if you don't have a valid driver's license, regardless of your race, that's going to take
more time on the part of the police. The Over-the-Rhine neighborhood--this is the site of the--of
the riots several years ago, 9% of the Black drivers were stopped in the Over-the-Rhine
neighborhood while 5% of the White drivers were stopped in the Over-the-Rhine. Now, policing
might just be different. Over-the-Rhine is a high crime neighborhood or it certainly
was during this period and policing was very different. And you--and those police might
have asked a lot more questions if anyone stopped in that neighborhood. Ninety-three
percent of Black drivers were Cincinnati residents, whereas 61% of the White drivers were Cincinnati
residents. Now you might wonder whether the Cincinnati Police should be treating the Kentucky
drivers differently from the Cincinnati drivers, but it's possible that there are factors other
than race at play here. All right, so now that we know all of these differences, what
do we do about it? If we see that--we don't know now whether that difference between 55%
and 65% is due to race or it's due to one of these other factors that's listed in this
table. Yes? >> Can you just clarify the percentage number
>> RIDGEWAY: You got it. >> [INDISTINCT]
>> RIDGEWAY: No, the other way around. Of those--of the Black drivers stopped, 9% were
stopped in Over-the-Rhine. Okay. And you go look at, you know, one of the other ones,
you know, the I-75, you know, only--of the Black drivers, 4% were stopped on I-75 on
the major freeways there while 11% of the White drivers. So, the white drivers are just--they
have--they're stopped in context. But the kinds of stops that occur on the freeways
are different from those that occur inside the city and that may or may not impact the
length of a stop, but it's fact that we see that are different between Black and White.
So let's try to get--so, this comparison now is more than just comparing of Black and White,
it's comparing Black and White in different context as in apples to oranges comparison.
So let's try to get at an apples-to-apples comparison. So what I did is identified non-Black
drivers that were stopped at the same time, same place, same context as the Black drivers
in Cincinnati. So have a look. Now, we've got stops involving White drivers where 20%
of them had invalid driver's licenses, 10% of them were stopped in Over-the-Rhine, 92%
of them were Cincinnati residents. So in all respects within one or two percentage points,
we've got White drivers that were stopped at the same time, same place, same context
as the Black drivers, okay? So any difference we observe between these 27,000 Black drivers
and these 5,000 White drivers, that's not due to time, place, context. That's more likely
due to race. So now let's see what the results are there. Of those--of those White drivers,
57% of them had stops lasting less 10 minutes, okay? So much of the difference between the
55% and the 65% has to do with factors other than race. It has to with time, place, context.
Now, there is still a difference, but the difference is 55% and 57%. That's still a
difference. And if you're the Black driver that has a stop lasting excessively long,
this is no--this is no consolation. But the fact is this is the number that the public
needs to worry about and discuss and debate. If you look at search rates, another outcome,
and this is over several years, we see that searches of non-Black drivers vary between
about 2.6% and 3% of stops of White drivers result in searches. On the other end, here
are--look these white dots our here, the stops involving Black drivers are out here at more
like 6%, 7%, okay? So on the face of it, we've got large differences between the search rates
of Black drivers and the search rates of White drivers. A lot more searching of Black drivers
is occurring in Cincinnati. Now, let's do the same thing again. These stops occurred
different time, place and context. Let's match them up. Let's find--there's a Black driver
stopped in this neighborhood at this time. Let's find a White driver stopped at the same
time, place, context and check out the search rates. And that's the comparison between the
blue dot and the white dot. And if you noticed in the most recent year, I have data on 2008,
the White driver stopped at the same time, place, context or actually searched more than
the Black drivers in those neighborhoods. In fact, that pattern is consistent for the
last three years, 2006 through 2008. Yes? >> [INDISTINCT]
>> RIDGEWAY: I think there is. Although there have been some theories that the--of--what's
known as de-policing, that the police will back off when faced with, you know, with court
challenges and civil rights accusations, which might not be a bad thing in the end for improving
police community relations, improving public safety and so on. So it could be that, you
encountered a Black driver and maybe this cop normally would have done a search or anything
and now given, you know, the scrutiny, you know, backs off some. And we don't--we don't
have evidence of this. We don't have this documented. There are people that sort of
talk about this sort of thing happening, but it's not necessarily a bad thing. Yeah?
>> When you're picking a batch of drivers, how do you pick when you [INDISTINCT]?
>> RIDGEWAY: All right. So I was asked how do I pick the matches to make sure they're
sort of within the same context. I use the finest data that I have available. So in Cincinnati,
what I had was the neighborhood level. They've sliced up Cincinnati into, I think, 53 neighborhoods.
Most of them are fairly small that we would feel like it's called--it really is a neighborhood.
And so the best level I could match on neighborhood. In the New York example, you saw that it actually
had X-Y coordinates. And so I matched the entire distribution of X-Y--the distribution
of the X-Y coordinates to make sure that those overlap. Same with time, I'm matching the
continuous distribution of time, making sure that those are as close as I can get. So, in closing, I've gone through three examples
noting that it's not, in the end, that hard to do good analysis of the racial profiling
issue. Now, there's a lot of naive analysis out there. I mentioned the census comparison.
I've mentioned the sort of naïve comparison of let's compare the search rate for Whites
and search rate for Blacks and any differences there must be racial profiling. So that--I've
sort of dismissed those as reasonable analysis. But I've also shown that it doesn't take that
much effort to push it just a little bit farther to do real transparent calculation that the
apples to apples comparison match, Black and White drivers stopped in the same context,
matched officers with other officers that are patrolling the same time, place and context.
And when I present this to whether it's a police union or a community group, it's done
in just as much as you've seen it here. It's very transparent and easy to understand. There's
nothing that technical. But under the hood, there's a lot of statistics work to try to
make sure the matches look good and things like that. But in the end, the tables that
I've shown you say it all. And if you want more detail on any of these studies, you can
visit the Center on Quality Policing website, that's cqp.rand.org. All of our reports are
always free, downloadable on the web as our--all of other RAND's other reports. This is not
unique just to our policing work. But you hop in and sort of dive in and learn more
about the research there. Thank you. >> And we're happy to take questions if there's
anyone. >> So I didn't catch your--first, thanks for
coming and for speaking. I didn't quite catch--if you looked at the data for--like, what are
the results of one of the traffic stops? Like, does--is there--is there always a citation,
sometimes citations, sometimes not? >> RIDGEWAY: All right. So what I talked--the
two examples I gave were--of outcomes were whether search occurs, whether the length
of the stop was less than 10 minutes. But we also looked at whether a citation occurred,
whether a warning occurred, whether an arrest occurred, whether the person was asked to
step out of the car. There's lots of other things that you can also tract, and it just
depends on what the police have good data on. And Cincinnati, being under a consent
decree with the--as a result to this class action lawsuit, actually had really good data.
That they had a really good auditing process, they had monitoring going on and they were
actually collecting very rich data. So, we did look at all those questions. Now, I noted
in these examples sort of looking at across the entire city, you know, generally didn't
find large race disparity issues. Now, it turns out it's not always the case. I did
some work in New York where we found some neighborhoods where there was no evidence
of race bias, but others where there were large differences. Even after there's matching
by time, place, context, you would find neighborhoods in New York where there were Black--pedestrians
were still frisked more, searched more, arrested more, had more use of force used against them
and there was no explanation, you know, that--no excuse that could be made other than, you
know--it wasn't time, place, context. It had to be something else and race bias was one
of the remaining candidates. >> Have you found that, in general as a trend,
that police departments are becoming much more, I don't know, amendable to or interested
in statistical analysis just beyond racial profiling but in general?
>> RIDGEWAY: There's a lot of variability. So there are--there are certainly some departments
that are doing a lot of data collection. I take the Chicago Police Department, for example,
as a good model here. They have an award-winning data warehouse called the CLEAR System. They
compile lots of data both on sort of operational aspects to attack the crime problem but also
on public complaints. So, all of those are tracked out really well. What's not happening
yet is a lot of strategic analysis based on that. They've got a lot of tactical information.
If you want to know about a particular address, you can pull in from lots of data sources,
lots of information about the problems at that particular address. But now, you want
to look at, say, across all addresses that are problems and look for patterns and that's
where it becomes more challenging. They've developed a predictive analytics group, which
is sort of a new thing in policing. And they are being able to forecast tomorrow, just
like the weather, "There's a strong chance of a--of a gang retaliation shooting in this
one block area. Let's move some resources in." And I think they're--that's sort of the
limit in where things are heading. So there are some departments that are very progressive
in this way. >> Two questions: how is you work funded and
second, how--what kind of reactions do you get from people, from the community and from
the police when you present this information? >> RIDGEWAY: Yes. So the funding issue. So
RAND, as a whole, gets a funding from a variety of sources; primarily the federal government
but also from foundations, from private donors and philanthropists, from--and contracts from
cities. This particular work, I've blended together several projects. So the city of
Oakland was a grant from the National Institute of Justice. Cincinnati was a contract with
the city that as part of this court settlement. So in some sense, we're reporting to the--to
the city, but also reporting to the federal monitor who's reporting to the federal judge.
New York was with the New York City police foundation where they were trying to get ahead
of the problem before it got worse. Now, the second question was the reaction. Wide ranged.
Those who are sort of on the frontlines of this get it. They think these are the right
questions and the right approaches. They appreciate the transparency of the methodology. So, I
presented this to police chiefs, police unions, community organizations like the ACLU and
they get it and they buy into it and they want to do more. So can we--so we found--didn't
find problems here, let's start looking over in this space and apply those methods there
to see if there's issues there. There are always exceptions to that. There is always--I've
been at several community meetings when someone looks at me and says, "White guy, blue eyes.
We're tired of white guys with blue eyes looking at this data. We need, you know, someone with
a real look." Now, the good thing about this is they can count the same numbers that I
did and still in the end they will count 55% are searched and 57% of the whites are searched.
So in the end, it's just counting. So, there's not really a lot of statistical tricks going
on here. And that's a good thing about this. >> Thanks for the presentation. I don't think
I caught the first part of--so this data reflects actual violations of crime when these individuals
are being pulled over or it's the genesis is more based solely on race of the data that
you're getting? >> RIDGEWAY: So, traffic stops are all traffic
stops. The--now, allegedly, they are stopped for a reason, for some traffic citation or,
you know, a suspect--wanted for suspect, but that's--they should be stopped for some--for
some reason; same with the pedestrians, the stops involving pedestrians. They--in New
York, in order to make a stop at a pedestrian, they have to be suspected of a penal code
violation. There is sort of a standard. If they just go up and talk to someone and say,
"Hey, how's it going?" you know, that's not the level that would show up in this data.
>> And then the second question is, how is the police department--how do they train the
officers? Are they--is this profiling in the broadest sense part of the training that they
have or do you find that this issue of racial profiling is just endemic and on-the-job course
of action? >> RIDGEWAY: No. I mean, I think the--the
official training or they train--try to train out racial bias. I mean--and, in fact, NYPD
has done some innovative programs of bringing people to, you know, tie a recruit classes
to Black neighborhoods and have that community sort of yell at them about all the problems
that they're--that they've had. So the police--that the new recruits can sort of understand the
abuses that have happened in this--in this neighborhood. So that when they're policing
that neighborhood, this doesn't--they don't police in the same way of the past. Now, there
are sort of--there are certain cases where things go wrong and the officers, you know,
abuse their police powers for racial reasons, for non-racial reasons. And so, I don't think
the training--the anti--racial profiling training is up to the level that needs to happen nor
is it apply--there are 17,000 law enforcement agencies across this country. There's a lot
of variability in the quality of training of any kind much less issues of racial profiling.
I think that's still a work in progress. Yes? >> I had a comment about your traffic stop
study where you said that it's much more difficult to tell if somebody is Black or White, you
know, after dark. >> RIDGEWAY: Yes.
>> But I think the cars probably are similar. I mean, you can--you can kind of go with this
whole, like, you know, Black people in a car potentially have a different type of car,
different modifications, whatever. >> RIDGEWAY: Yes, that's--so that--yes. This
issue doesn't bias the test but it undercuts the statistical power of the test in the same
way if the city was lit up at night, you know. It doesn't undercut the--it doesn't bias the
test that were--the statistical test but it under-powers it because some fraction of those--some--at
the nighttime stops, half the stops are still lit up like they are during the day and half
of them are dark like we need them to. It's essentially an instrumental variable analysis
and things like the--using the car type as a proxy or street lighting undercut the quality
of the instrument to separate--to predict visibility. So, indeed, if car type was completely
correlating with the race of the driver, then the instrument is no longer valid and it's
just--you always get a no results out of this. But we got to assume that there are at least
some drivers in a Honda Accord that are White and some other drivers in a Honda Accord that
are Black and even though... >> You don't take that into account?
>> RIDGEWAY: No. There's no need--there's no need to because the mix of cars on the
road won't change between when it's light out and when it's dark out. So there isn't
a need to do that for reasons of bias. But the only thing that you could do is to improve
the power is--you know, and then you start having to eliminate intersections that are
well lit. From the analysis, things like that, you know, might improve the power of the--of
the--statistical power of the test. >> You stated your control for context where,
when and what you might--what time of the day it was.
>> RIDGEWAY: Right. >> Do you also control for there possibilities
that certain categories would violate rules more often? Because that would--then that
would disprove that there's actually racial bias. You just--just police doing its work
if the car is speeding and if, say, tall people speed more, then it's not bias against all
people, just tall people speeding more. Do you have control for that?
>> RIDGEWAY: Yes. I'm not in the decision to make a stop. That part is really hard,
I think. In the--the first part, you know, a decision to make a stop, I've--because there's
a lot of police discretion in who to stop and who not to stop. And there's days that
say, if a cop follows you for a mile, he'll find some reason to stop anyone. So in the
other parts, I'm dealing with stops--police activity that happened after the stop takes
place. So the decision to search, decision to write a citation and things like that.
And there, we sort of at least--they've--the stop has met the level of--the police have--it's
past the police's discretion that this is someone I needed to stop for some reason.
Now if we start finding differences in the search rates that, say, the search rates of
Blacks are much lower than search rates of Whites, then you might wonder maybe a lot
of those stops shouldn't have happened in the first place because they're stopping people
and going, "Oh, this is the wrong guy" or "We didn't really need to stop this person."
So if we found large differences in search rates or large differences in citation rates,
that would lead you to that conclusion. Now, we're not seeing those large differences.
We're seeing smaller differences, differences nonetheless. So, there might be some of that
going on but not at a--not at a large magnitude. >> You also have another thing where--you
also have another thing going on with trying to figure out who to stop is that you have
to measure who you didn't stop and that's pretty hard.
>> RIDGEWAY: Yes. So the--some of the original--some of the other studies--well, I actually have,
like, you know, graduate student staying on clipboards--I mean, they're staying with clipboards
the intersections, watch cars go by much like the New Jersey study and say, "Okay, oh that's,
you know, white guy, white guy, you know, black guy," and trying to record this down.
But then, not everybody that passes by is at risk for being stopped by a cop. And so
then they try to say, "Well, let's make this a little bit more difficult. Let's put a radar
gun" and say, "All right, let's just get the race distribution of who's exceeding the speed
by--speed limit by five miles per hour." And then, well, cops aren't stopping. Everyone
that's over five miles an hour, they're looking for other things, you know, brake lights out,
seatbelt violations and stuff. And so the list of things that you try to get the graduate
student to record gets difficult. Now, the good thing about the Daylight Savings Time
analysis I described is you don't have to worry about that. We just have to assume police
practice is the same, you know, at 6:30 regardless of which side of Daylight Savings Time you're
on. So they can do whatever it is that they normally do, whoever they usually stop. But
we get to--we skip all--we get to skip all that.
>> Actually, I have two questions. I'll just ask one now, but what--you said that police
practice is the same where--regardless of whether it's day or night at 6:30. I was just
wondering if that's actually--you stated that kind of as an assumption, but is that--is
that true? >> RIDGEWAY: Yes. Well, I mean, with the exception,
we do drop, like, headlight violations, for example, because during--those only happen
when it's--when it's dark out and there might be differences in race of whether, you know--differences
in whether you have headlight violation between Black and White. So, we dropped some things
like that. Now, I've talked to the police saying, you know, "What happens when it's
dark out?" And they're like, "Oh, no. We're on a strict shift basis so we don't reallocate
our force. You know, our shift ends at whatever 9 o'clock whether it's before or after Daylight
Savings Time." So our allocation is the same. And in the end, they look for the same sorts
of things; you're running red lights, going through stop signs, speeding, things like
that. >> Yes. I mean, I actually think it's a reasonable
assumption. I'm just pointing it out because it is an assumption and...
>> RIDGEWAY: Yes. >> ...you know, it could be questioned, you
know. Hypothetically, maybe there are people that are only racially biased during the daytime
or nighttime. >> RIDGEWAY: That would completely mess up
this analysis. >> Yes.
>> RIDGEWAY: That's right. >> Some people become absurd.
>> RIDGEWAY: Yes. When it gets dark, I turn into a racial profiler.
>> Can I just make a brief comment on that? Because if you remember the Stanford Prison
Experiment that you may be familiar with where a Stanford psychology professor had...
>> RIDGEWAY: Oh, yes. >> ...I think undergrad students--doesn't
matter undergrad or grad. But one group was prisoners, one group was guards and then inevitably,
I guess, the guard started torturing the prisoners. And, actually, the people who wanted to commit
these atrocities in this experiment did prefer the night shift and they did act that during
the nightshift. >> RIDGEWAY: Yes. But we have to remember
though that it's the same--it's the same people--or 6:30 is not necessarily the night shift, but--or
they're--it's the same people that are going to be there when 6:30 is light out and when
6:30 is dark. So it is--it'd be--now, the first analysis, the naïve analysis that just
compared daylight to darkness, we're talking about different people, the people that are
getting trouble with the police while you're sent to the midnight shift, you know. That--those
sorts of things are problem with the naïve just daylight/darkness analysis. That's why
we go to really slice it up by clock time. Yes.
>> How do you determine, like, at what locations you're actually collecting the data from and
does the location either validate or disvalidate the results that you're getting? I mean, to
me, it seems like if you're in a predominantly Latin community that you're going to probably
have more Latin population violating a certain rule or wherever ethnicity that you find yourself.
So, how do you go about? >> RIDGEWAY: So I get--I get all the data.
All police activity throughout--across the city. So for New York City, no matter, you
know, if there's a stop that takes place, I want the data on it. So there's no selection
on that. Now, if I'm looking at search--if there's a concern about the high level of--the
search rate of Black pedestrians in New York City, I want to look at the distribution across
the city of where stops of Black pedestrians occur. And they're different from where stops
of Latinos and stops of Whites occur and stops of Asian occur. But if we're concerned about
the search rate of black pedestrians, we've looked at the distribution of where they occur.
And let's find stops of whites that occur in those same time, place, context and match
those up to the same distribution. These other methods, that I'm not such a fan of, do have
to pick an intersection to station a graduate student to write down a clipboard and selecting
those--if you do it, there are good ways--just good ways of doing that just like you would
standard statistical design of experiments, you know, randomly picking our sections things
like that. You can do it well and you can also do it poorly.
>> I came across some--a nice piece of work where some police departments are scheduling
shifts based on the weather forecast because they've got the correlations between what
kinds of things a police need to do as a function of actual instantaneous weather so they can
schedule shifts the next few days based off the weather forecast. I can see a real financial
benefit to them doing that because the--that would keep their stuffing level lower. So
the data collection that tracks stuff as a function of weather is completely offset.
How a traditional collection effort is there for the police force to get that from that
level of collection is going to direct financial benefit through to this one which has a would-you-see-a-risk-of-something-getting-really-irritated-with-you
benefit. >> RIDGEWAY: So, here, there is these lawsuits.
So that's it. >> Yes. But I mean, how--what's the collection
effort for the police force to provide you with the data set that you can analyze?
>> RIDGEWAY: Well, I'm not sure I understand the question. I mean, a lot of the departments
aren't collecting anything on their interactions with the public. There was about 14 to 15
states that require all the police departments to collect data on all stops, so they're going
to do it anyway. California, it's not a requirement for police to document interactions with the
public. Some cities, the first one was San Jose, although I don't know if they're still
doing it, they did it sort of voluntarily because they thought it would be a good thing
to track for just--because that was part of their business. You track things that are
at your core business. L.A. did it because of lawsuit. So, now, I don't know if these
departments would do it because of--for necessarily for business. Not all--so a lot of departments
will say it's just a waste of money and officer time to collect this data that is never going
to be looked at or if it's looked at it's going to be looked at poorly. So that's certainly
a complaint that I--that I hear. >> I guess a follow-up question or phrasing
that in a different way might be are there policy implications for states in the federal
government for this kind of data collection or would it really be that to have a blank
collection of data would be just way too much overkill and it's better to target areas where
there--where people believe there's a problem, where there are a lawsuits and so on?
>> RIDGEWAY: Yes. >> So.
>> RIDGEWAY: So, there's a--some federal legislation called the End of Racial Profiling Act. It's
been introduced in the Congress just about every term for the last four terms and most
recently in 2010. And there's sort of four pieces to this. I'm not sure I remember all
of them, but essentially it bans the practice of racial profiling in the United States.
That's actually not been clearly stated by the federal government prior to this. The--and
then the other part is to require all--any police department receiving federal funds
to collect data on traffic stops and report on it. So, there is sort of this movement
to make all the departments do it. I'm not sure that's necessarily cost beneficial. Any
time the justice department sort of--the justice department through Section 14141 of--this
is the power that gives the justice department the authority to investigate, bring lawsuits
against police departments that are--have suspected patterns or practice of civil rights
violations. So when those come--when those sorts have come up, the first thing they do
is they're going to start doing data collection and it's not just on traffic stops; it's on
any sort of public contact, it's about any arrest you make, it's about any use of force
you make, any use of dogs. Start collecting data. So the first line of business that they
are required the first stage. And that seems like a reasonable practice. And--but I also
don't think it necessarily has to be the U.S. Department of Justice that makes that happen.
Any community that is concerned about the--has suspicions of its police department or feels
that there's a deterioration of police community relations, one of the first things that that
police department can do is we're going to start collecting data and monitoring this
and report to the public in a transparent way to make sure we don't get to the point
where the Department of Justice is knocking on our door doing an investigation or filing
a lawsuit. >> Is there action for [INDISTINCT]?
>> RIDGEWAY: Yes. >> [INDISTINCT] from the officers who are
[INDISTINCT]. >> RIDGEWAY: Yes, that's a good question.
It's actually not too hard to validate at this. So, for example, in Cincinnati, when
they stop someone, they call in to dispatch, you know, "Take me out of the pool because
I'm--for the next 10 or 15 minutes, I'm going to be dealing with this traffic stop." And
so, we get a record from the dispatch this guy was engaged in a traffic stop and we look
for the corresponding form. And that at least gets to the existence of a traffic stop taking
place. Now, whether it's what's on the form, whether the search actually took place or
didn't take place, that's not necessarily--that's harder to audit. What certainly happens is
if the person comes in with a complaint and says, "I was searched" and the form says they
weren't searched and then they got the in-car camera that shows the officer searching but
it's not on the form, that's a big problem and every officer knows that it's not worth,
you know, fudging the form because this sort of thing can happen.
>> Okay, one last question. >> Okay, one last.
>> RIDGEWAY: I'll stick around if you want to ask more.
>> A quick follow up to that is just that because earlier you talked about when they
stopped pedestrians and they just talked to them and say well, it's not a stop so they
don't do the data collection. Same thing here, if they--if they were to stop someone and
not call dispatch, then they don't have a record of it. Is it possible, you know, that
it's an example of harassment or something that's not documented?
>> RIDGEWAY: So if--if it--you are now talking the difference between pedestrian stops and
a traffic stop. So traffic stop... >> Well, [INDISTINCT] the traffic stop as
well where they stop someone but don't call dispatch.
>> RIDGEWAY: They... >> Hypothetically.
>> RIDGEWAY: It could vary by department. But in Cincinnati, they would call dispatch
because otherwise they're on call--like if a 911 call comes and they're out of their
car, you know, dealing with the traffic stop that's...
>> I'm not alleging that it's happening. I'm saying that hypothetically, a policeman can
stop someone, get out of their car and not call dispatch is the example of that.
>> RIDGEWAY: Yes. Yes. >> Okay. And--but the other question I wanted
to ask had more to do with earlier when you're talking about in Oakland how in Oakland Hills
they don't have as many police... >> RIDGEWAY: Yes.
>> ...presumably because there's not as much crime in Oakland Hills, but you're not alleging
this is happening. But, like, is there--there's also the potential that when you have more
police in an area, you're likely to find more crime. So for the example of speeding tickets,
if you had more police in Oakland Hills... >> RIDGEWAY: Yes.
>> ...more people would be getting speeding tickets. But you don't put police there because
speeding tickets are not considered to be like as important as concentrating--like looking
for other types of crime. So I guess what I'm getting to, is there...
>> RIDGEWAY: I get it. I get it. >> Is it the potential of racial bias and
how do you distribute police? >> RIDGEWAY: Yes. So, I mean, that is a legitimate
concern that the reason why there's such despair is you've got 10 times more police officers
in a Black neighborhood than in a White neighborhood. Now, this is different from racial profiling
that is the individual officers are targeting Black residents, Black drivers, Black pedestrians,
but it's a--it's a very different problem. And the good thing is it has an easy solution.
It just involves moving officers somewhere out and that's something that can be negotiated
with the community, with the police department and with the--with the political bodies in
that--in that community, but it's a very different problem for racial profiling. So, New York
could solve many of its racial disparities simply by reallocating a lot more officers
to, say, Manhattan South. But then, what you'll find is, you know, next year, there'll be
complaints about the response times to calls in Queens, so. And you'll find biases there
next year that you didn't find this year as a result of that. So there this is balance
between getting the allocation that the community wants and--but that doesn't sort of exacerbate
racial disparities that you're most concerned about like response time.
>> [INDISTINCT] control for that without actually doing what is probably the wrong thing which
is you put things where you think that there's less crime. The livelihood [INDISTINCT]
>> RIDGEWAY: What do you mean control for it?
>> Control--is there a way to excessively control that?
>> RIDGEWAY: I don't want to control for it. I'm going to--I want to--because this is the
way the police are functioning right now, so I'm going to--and I'm trying to detect
whether individual officers or department as a whole is targeting minorities in their
community. >> Is there a way to detect a racial bias
that [INDISTINCT]. >> RIDGEWAY: So, right. So now you're saying
like is the--does the allocation of police officers make sense in some way. And there's
a lot of different ways that police reallocated. Some have, you know, their computer programs.
There's companies that's, you know, developed allocation models just based on your input,
the points of your stations and things like that and where your calls are coming from
and it says--and it does all your staffing and things like that and it's race blind so--because
the computer just knows locations of cars and resources and demand for service. And
so that's a safe way of doing it. But then, politicians will also weigh in. And, usually,
it's--every community turns out wants more police service in their area. They might not
like their--you know, the quality they're getting, but most communities--the political
leadership will be arguing for more police in their neighborhood rather than in that
other neighborhood that doesn't need it so much. Big Bell in Chicago about this.
>> I think we're out of time, so. >> RIDGEWAY: Okay.
>> Thank you, Greg, for coming. >> RIDGEWAY: Sure. Glad to be here.