The Crime Reduction Effects of Public CCTV Cameras: A Multi-Method Spatial Approach

Abstract

Public Closed Circuit TeleVision (CCTV) initiatives have been utilized as methods of monitoring public space for over two decades. Evaluations of these efforts to reduce crime have been mixed. Furthermore, there has been a paucity of rigorous evaluations of cameras located in the USA. In this analysis, crime in the viewshed of publicly funded CCTV cameras in Philadelphia, PA, is examined using two evaluation techniques: hierarchical linear modeling and weighted displacement quotients. An analysis that incorporates controls for long-term trends and seasonality finds that the introduction of cameras is associated with a 13% reduction in crime. The evaluation suggests that while there appears to be a general benefit to the cameras, there were as many sites that showed no benefit of camera presence as there were locations with a positive outcome on crime. The policy implications of these findings are discussed.

Full text

JUSTICE QUARTERLY VOLUME 26 NUMBER 4 (DECEMBER 2009)
ISSN 0741-8825 print/1745-9109 online/09/040746-25
© 2009 Academy of Criminal Justice Sciences
DOI: 10.1080/07418820902873852
The Crime Reduction Effects
of Public CCTV Cameras: A
Multi-Method Spatial Approach
Jerry H. Ratcliffe, Travis Taniguchi and
Ralph B. Taylor
Taylor and FrancisRJQY_A_387557.sgm10.1080/07418820902873852Justice Quarterly0741-8825 (print)/1745-9109 (online)Original Article2009Taylor & Francis00
000000002009JerryRatcliffejhr@temple.edu
Public Closed Circuit TeleVision (CCTV) initiatives have been utilized as methods
of monitoring public space for over two decades. Evaluations of these efforts to
reduce crime have been mixed. Furthermore, there has been a paucity of rigor-
ous evaluations of cameras located in the USA. In this analysis, crime in the
viewshed of publicly funded CCTV cameras in Philadelphia, PA, is examined
using two evaluation techniques: hierarchical linear modeling and weighted
displacement quotients. An analysis that incorporates controls for long-term
trends and seasonality finds that the introduction of cameras is associated with
a 13% reduction in crime. The evaluation suggests that while there appears to
be a general benefit to the cameras, there were as many sites that showed no
benefit of camera presence as there were locations with a positive outcome on
crime. The policy implications of these findings are discussed.
Keywords CCTV; video surveillance; repeated measures; hierarchical linear
modeling; weighted displacement quotients; Philadelphia
Introduction
Closed Circuit Television (CCTV) seeks to reduce crime primarily by increasing
the perception among potential offenders that there is an increased risk of
Jerry Ratcliffe is a Professor in the Department of Criminal Justice at Temple University. Recent
books have included Intelligence-Led Policing, and Strategic Thinking in Criminal Intelligence
(second edition). Further details are online at jratcliffe.net. Travis Taniguchi is a doctoral candidate
in the Department of Criminal Justice at Temple University. His publications can be found in Justice
Quarterly and Crime Patterns & Analysis. Ralph B. Taylor is a Professor in the Department of
Criminal Justice at Temple University. Research interests include incivilities, communities and
crime, reactions to crime, guns, juries, and DNA policies. Publications are listed online at
www.rbtaylor.net/pubs.htm. Correspondence to: Jerry H. Ratcliffe, Department of Criminal Justice,
Temple University, 1115 W. Berks Street, Philadelphia, PA 19122, USA. E-mail: jhr@temple.edu
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 747
detection and capture (Ratcliffe, 2006). Proponents of CCTV argue that
increased surveillance will reduce crime and increase arrests of offenders, while
opponents lament the invasion of privacy in public areas and point to research
findings that are less than definitive. Irrespective of the theoretical validity of
the crime prevention role of CCTV systems, the growth of CCTV schemes across
numerous countries (including the USA) has been substantial. The history of
CCTV technology is one of rapid evolution from static, low-resolution cameras
to high quality technology solutions that can pan, tilt, and zoom at the
command (through either wireless interface or fiber optic cable) of remote
operators connected to a police radio network. This implementation and tech-
nological expansion of CCTV schemes has, until recently, taken place in an envi-
ronment largely devoid of rigorous evaluation of the effectiveness of CCTV to
prevent crime (see Welsh & Farrington, 2007, for a comprehensive review).
The emergence of CCTV has taken place during a period when crime analysis
has improved in resolution (both spatial and temporal) enabling academic
research and practitioner focus to become more place-specific rather than
generalized to the neighborhood level (Mazerolle, Hurley, & Chamlin, 2002),
and during a period that has seen the rise of problem-oriented policing (Clarke,
2004; Goldstein, 2003) as an operational strategy that can capitalize on the
analytical improvements available. Intelligence-led policing has provided a busi-
ness model to coordinate crime detection and prevention activities, through
which problem-oriented policing can flow in an operational environment (Ratc-
liffe, 2008). From the earliest wide-scale implementation of CCTV technology in
Britain in the 1980s, however, the assessment of CCTV schemes has been signif-
icantly hampered on two fronts. Numerous well-meaning evaluations have
lacked either an impartial perspective and/or methodological rigor. For exam-
ple, many have been either conducted by city agencies or technology companies
involved in the scheme and whom may be perceived to have vested interests in
the evaluation outcomes; or, like the earliest independent evaluation of a CCTV
implementation from King’s Lynn, UK (Brown, 1995)—where 19 cameras were
installed at public car parks across the city—had methodological limitations due
to a lack of controls for seasonality or long-term temporal trends. Numerous
other studies since King’s Lynn have lacked measures of control areas, controls
for seasonal variation, or have been absent of any indicators of potential
displacement (or diffusion of benefits).
Furthermore, the existing evaluation literature demonstrates considerable
variation in not only methodology, but also outcome measures and independent
variables. Some studies examine the impact of cameras on crime within a
defined distance of CCTV cameras (Harada, Yonezato, Suzuki, Shimada, Era, &
Saito, 2004), while others surveyed residents in camera areas for their percep-
tions of how crime has changed (Squires, 2003). Other studies have interviewed
key stakeholders (Hood, 2003) or examined emergency room attendance levels
related to assaults (Sivarajasingam, Shepherd, & Matthews, 2003).
Welsh and Farrington’s (2007) systematic review identified four essential
criteria for inclusion in their study:
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

748 RATCLIFFE ET AL.
(1) CCTV was the main intervention examined;
(2) the outcome measure was crime;
(3) the evaluation had a minimum methodological design that incorporated at
least before-and-after measures of crime in experimental and comparable
control areas; and
(4) there was a minimum number of crimes (20) recorded in the experimental
area prior to the CCTV implementation.
Even within this basic evaluation criteria framework, Welsh and Farrington
(2007) had to exclude numerous studies because they lacked appropriate and
comparable control areas. They were able to identify 44 relevant studies, but
only 22 that were applicable to city centers and public urban areas. Of these 22
studies, 17 were conducted in the UK, two in Scandinavia, and three from the
USA; however, the three from the USA were actually at three locations in
Cincinnati, Ohio, and were reported in a single journal article (Mazerolle et al.,
2002).
1
Given the scarcity of CCTV evaluations in the USA, the study conducted by
Mazerolle et al. (2002) is worth discussing in some depth both because of its
innovations and its limitations. This study utilized a multi-method approach to
evaluating the CCTV initiative in Cincinnati, Ohio. First, they used an innovative
method to measure both pro- and anti-social behavior in the vicinity of the
camera locations. The researchers reviewed random samples of video captured
by the cameras and coded activity along a number of dimensions including: the
quality of image being recorded, the number of people and vehicles at the
target site, and a range of pro-social (e.g., pedestrian traffic and people shop-
ping) and anti-social (such as people loitering, dealing drugs, or begging) behav-
iors. Use of interrupted time series models found a complicated relationship
between the implementation of cameras and their crime deterrence effects.
The second evaluation method evaluated the change in calls for service (not
recorded crime) in the areas surrounding CCTV target sites. Overall, it was
concluded that the implementation of CCTV systems produced an initial deter-
rence effect in the one to two months following implementation. This crime
suppressing effect, however, seemed to decline as people adapted to camera
placement. While the study conducted by Lorraine Mazerolle and colleagues is
methodologically sound and empirically sophisticated, there are a number of
limitations that are worth addressing. First, the use of calls for service rather
than recorded crime basically means that there has not been a single study of
CCTV in America that uses recorded crime as an outcome measure in a manner
that satisfies Welsh and Farrington’s (2007) systematic review criteria. Second,
the authors utilized circular buffers surrounding the camera target areas. These
uniform buffers are not sensitive to the actual viewsheds of the cameras. It is
possible that these buffers are both over-inclusive and under-inclusive of areas
1. There have also been two US studies that examined CCTV in public housing complexes. These are
discussed in Welsh and Farrington (2007).
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 749
where cameras may have a crime deterrent effect. Finally, Mazerolle et al
(2002) do not consider the possible displacement effects of camera implementa-
tion. The current evaluation attempts to rectify these limitations.
The lack of evaluation research seems a significant omission on the part of
the research community given the growing enthusiasm across America for CCTV.
While there are no national estimates on the extent of CCTV across America,
newspaper accounts suggest that CCTV cameras are being implemented at a
significant rate. For example, San Francisco has spent close to $1m on 74
cameras at 25 locations, and a further 25 cameras are planned (Bulwa, 2008);
and Washington, DC plans a $4.5 million expansion of its surveillance system
(Klein, 2008). This rapid and unprecedented expansion of video surveillance
technology is not just limited to the major urban areas (Welsh & Farrington,
2007). Reductions in technology cost and a perception that CCTV is a cost-
effective crime prevention tool, have driven investment in video surveillance in
municipal areas across America.
For all this enthusiasm for video surveillance, there has been a lack of high
quality, independent evaluation studies (Eck, 1997). Using Hierarchical Linear
Modeling (HLM) and Weighed Displacement Quotient (WDQ) methodologies, we
explore serious crime, disorder crime, and an all-crime measure combining serious
crime and disorder, with a multi-method spatio-temporal evaluation of 18 pilot
CCTV cameras across 10 sites in Philadelphia, PA.
Data
Camera types, locations, and implementation details
The Philadelphia pilot project employed two different camera types. Eight pan,
tilt, zoom (PTZ) cameras were installed between July 2006 and October 2006.
These cameras have the capacity to tilt up and down, pan around the surround-
ing area, and zoom. Examination of the zoom capacity by the researchers indi-
cated that the camera allows the police officer to read a car license plate more
than a block away, and observe street activity up to three blocks distance, if
the view is unobstructed. The video feed is routed directly to police headquar-
ters where a police officer monitors all PTZ cameras in real time. The images
are also recorded digitally, with a hard drive storage capacity sufficient to store
images for 12 days.
The remaining 10 cameras did not allow for live monitoring. The Portable
Overt Digital Surveillance System (PODSS) cameras, as implemented in Philadel-
phia, provide a moveable, self-contained digital camera and recording system
that is housed in a bullet-resistant unit with flashing strobe lights to draw the
attention of the public and potential offenders. The cameras are visually quite
different from the PTZ cameras, being housed in a much larger and more visible
unit. As implemented in Philadelphia, these cameras are not monitored at
police headquarters; however, nearby patrol officers with the correct
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

750 RATCLIFFE ET AL.
equipment in two-officer cars can theoretically view the feed from cameras
over a wireless link (though the senior officer in charge of implementation did
not believe this took place on any regular basis). The system is also able to
record up to five days of street activity on a digital hard drive. When a crime is
suspected to have occurred within the view of the camera, a police officer
meets street engineering personnel from the city and the digital video record
hard drive is retrieved from the unit with the aid of a crane. Discussions with
police officers suggested that the whole process can take up to two hours.
Although there are a total of 18 cameras at 12 locations, some cameras are
located so close to other cameras (within a block’s distance) that the decision
Figure 1 Pilot project camera locations in Philadelphia, PA.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 751
was made to evaluate spatial sites rather than individual cameras. The situation
is further complicated by cameras three to five which are not located at a single
site but are close enough that any potential displacement would likely be to
overlapping areas. Therefore, we collapsed sites three to five into a single loca-
tion leaving eight different evaluation areas, labeled 1, 2, 3–5, 6, 7, 8, 9, 10 on
Figure 1. The choice of camera was dictated by the locational demands of the
area, and was made by Philadelphia Police Department (PPD) officers prior to
our involvement in the project.
Figure 1 Pilot project camera locations in Philadelphia, PA.
Offender perception or camera viewshed?
It can be argued (e.g., Ratcliffe, 2006) that from a rational choice perspective
(Clarke & Felson, 1993; Cornish & Clarke, 1986; Jeffery & Zahm, 1993) comes
the supposition that cameras may work to prevent crime if two criteria are met:
the offender is aware that the camera may be monitoring their activity, and the
offender perceives that the risk of capture by police may outweigh the benefits
of the crime they are considering. As crime prevention is therefore a feature of
offender perception, it may be that irrespective of whether cameras may only
be able to see a certain amount of public space, offenders perceive that the
cameras can observe their activity to a greater or lesser range. The choice from
an evaluation standpoint is therefore to define target evaluation areas based on
possible offender perception of camera range, or on the actual area that the
camera can view.
The difficulty with offender perceptions is that they are not measurable
without extensive and expensive interviewing. Furthermore, the resultant
offender perception will most likely vary from person to person. In other words,
while the range of a CCTV camera—as perceived by a criminal—is in the eye of
the beholder, finding and interviewing suitable beholders is beyond the budget
of most studies, and the results are likely to be quite variable.
The second option is to define the boundaries of a likely impact area by the
extent of actual camera vision. The advantages of this approach include being
able to: work with camera operators to establish the viewshed of cameras;
incorporate the natural constraints on viewsheds (such as trees or buildings);
include areas that camera operators are likely to initiative action within; and
build spatial map units that reflect a single areal unit for the camera. In the
analysis conducted here, this approach is taken; namely, to map the actual area
that cameras can view. Furthermore, in the event of a misspecification of the
surveillance zone, the weighted displacement quotient (WDQ) analysis is able to
incorporate a measure for diffusion or displacement.
For the PTZ cameras, we visited the CCTV viewing station at PPD headquar-
ters. The researchers worked with PPD officers to map the individual viewsheds
of the cameras by panning and zooming the cameras and discussing active view-
ing areas with the officers. In combination with street maps, we were able to
establish the workable range of each camera. This approach is sensitive to the
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

752 RATCLIFFE ET AL.
geography of the camera location and was therefore deemed preferable to the
selection of an arbitrary buffer distance that counts crime up to a fixed distance
in all directions from the camera, a technique employed by Harada and
colleagues (2004). Because, as previously stated, the authors were unable to
view the video feed from PODSS cameras (it is not a live feed in Philadelphia),
the target area was designated as simply the junction (street intersection)
where the camera was located, a choice supported by PPD officers that had
viewed footage from fixed cameras.
Two areas were designated around each camera site. The first area was
designated the target area—the area where the cameras are expected to have a
positive effect. Buffer areas were also generated around camera locations.
These areas were designed to be likely places in the surrounding neighborhood
of the cameras where crime activity could potentially be displaced. The buffer
area is also a zone where potential diffusion of benefits (Clarke & Weisburd,
1994) could occur. This can happen when the cameras exert a benefit to
surrounding areas beyond their target area, and may occur because offenders
move out of the general area of the camera, or offenders at unviewed areas
curtail their activity because they think the camera can still see them. Displace-
ment areas began as simple 500 ft buffers surrounding the target areas. 500 ft
was chosen as a rounded median estimation of the length of one city block.
These 500 ft buffers were then adjusted to account for local geography and road
patterns surrounding each location. This means that at some locations the
displacement buffer was slightly less than 500 ft while in others it was greater.
While it may seem intuitively better to have uniform displacement areas, doing
so ignores the substantial variability in the geography surrounding camera
implementation areas. For example, the use of actual camera viewsheds can
mean that a 500 ft buffer stretches to just short of a neighboring intersection.
In circumstances like this, the addition of an extra 20 ft is sufficient to include
the street intersection (and thus the crime at that location) and create a buffer
that is a more realistic approximation of the likely displacement area. The
method utilized here, while requiring more effort and a greater understanding
of local geographic conditions, produces more realistic target and displacement
areas. Figure 2 illustrates the unique shape of PTZ buffers compared to the
more traditional buffer approach that was utilized for the PODSS cameras.
Figure 2 Example of PTZ and PODSS camera target and displacement area. The camera location is shown as a small cross in the center of each image, while the lighter (hash-marked) area near the center is the designated target area. Potential buffer areas (to assess displacement or diffusion of benefits) are shown as dark grey. Individual lines indicate a road network.
Finally, as a control on general trends in the surrounding areas beyond the
target and buffer area, the surrounding police district(s) beyond the displace-
ment areas became designated as a control area. As such, the control areas are
comparable across both socio-economic and organizational parameters. First,
the control areas represented territory that was within the region of the target
(experimental) area, so remained within the generalized socio-economic struc-
ture of the block of neighborhoods that make up the police district. By being
within the same police district as the cameras, the control areas were also
susceptible to the same organizational forces that affected the camera areas.
Researchers using control areas that are in entirely different police districts run
the risk that the districts without cameras may conduct a different style of
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 753
policing or introduce their own crime fighting initiatives that could confound
the study. Using control areas that are in the same district as the cameras
means that both target and control areas are policed in generally the same
manner, resulting in greater comparability of area for the study.
Where a camera target or buffer area intersected more than one police
district, we used the remainder (all areas not included with target and buffer
areas) of the intersecting police districts combined. Table 1 provides details of
how these control areas were constructed, along with the number of months the
cameras had been operational at the site at the time of evaluation, the specific
dates of the pre-/post-camera implementation evaluation period, the camera
type, the number of cameras deployed at that location, and the police district.
As can be seen from the last column of the table, the definition of control areas
was confounded by camera sites being relatively close to each other, and some-
times on the boundary between police districts.
Crime data
Crime data from January 2005 through August 2007 (32 months) was sourced
from the PPD’s Crime Analysis and Mapping Unit. This dataset contained infor-
mation about crime type, date, and the x-y coordinates of the crime location,
as geocoded by the PPD to a successful geocoding hit rate in excess of 97%; a
satisfactory level in excess of minimum geocoding levels estimated through
simulation processes (Ratcliffe, 2004).
This evaluation limits crimes examined in the study to those that could be
expected to be influenced by CCTV cameras. Therefore, only crimes that gener-
ally occur on the street were included in the analysis. In other words, theft from
a vehicle is included while theft by shoplifting (because it would happen inside a
store away from the view of the camera) is not. The crimes are aggregated into
Figure 2 Example of PTZ and PODSS camera target and displacement area. The
camera location is shown as a small cross in the center of each image, while the light-
er (hash-marked) area near the center is the designated target area. Potential
buffer areas (to assess displacement or diffusion of benefits) are shown as dark grey.
Individual lines indicate a road network.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

754 RATCLIFFE ET AL.
Table 1 Specification of weighted displacement quotient parameters
Site
Months
camera
operational
Pre-
implementation
dates
Post-
implementation
dates
Camera
type
No. of
cameras
Location
(police
district) Control area composition
19Jan 2005–Nov
2006
Dec 2006–Aug
2007
PODSS 1 14 14th police district crime totals, minus target
and displacement areas for camera at site 1
29Jan 2005–Nov
2006
Dec 2006–Aug
2007
PODSS 2 39 39th police district crime totals, minus target
and displacement areas for cameras at site 2
3–5 10 Jan 2005–Oct
2006
Nov 2006–Aug
2007
PODSS 4 25 25th and 39th police district crime totals, minus
target and displacement areas for cameras at
sites 2–5
610Jan 2005–Oct
2006
Nov 2006–Aug
2007
PODSS 1 25 22nd, 25th and 26th police district crime totals,
minus target and displacement areas for
cameras at sites 3, 4, 5, 6, 7, and 9
711Jan 2005–Sep
2006
Oct 2006–Aug
2007
PTZ 2 22 22nd police district crime totals, minus target
and displacement areas for cameras at site 2
811Jan 2005–Sep
2006
Oct 2006–Aug
2007
PTZ 2 23 6th, 9th and 23rd police district crime totals,
minus target and displacement areas for
cameras at site 8
914Jan 2005–Jun
2006
July 2006–Aug
2007
PTZ 2 26 26th police district crime totals, minus target
and displacement areas for camera at site 9
10 11 Jan 2005–Sep
2006
Oct 2006–Aug
2007
PTZ 4 17 17th police district crime totals, minus target
and displacement areas for camera at site 10
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 755
three categories; serious crime (UCR Part 1 street offenses), disorder crime
(UCR Part 2 street offenses), and all crime (the sum of the unweighted serious
and disorder crime categories). The full list of crimes included in the analysis
can be found in Appendix A.
Analysis
Two methods were utilized to investigate the impact of camera implementation
upon localized crime. HLM allows for rigorous statistical evaluation of camera
implementation while controlling for factors such as seasonality and ongoing
trends. The use of HLM in this analysis, however, is limited because it is not
possible to investigate specific cameras or specific locations. In order to further
investigate the effects of specific cameras we utilize WDQ as described by
Bowers and Johnson (2003). We discuss the HLM model first.
Hierarchical Linear Modeling
HLM is a type of statistical analysis that recognizes nested data structures
(Raudenbush & Bryk, 2002; Snijders & Bosker, 1999). This also applies to
repeated observations across individuals or locations (Laird & Ware, 1982). The
current analysis examines time nested within camera locations. The particular
analysis completed has a number of practical benefits. First, it includes a vari-
able that statistically controls for seasonal effects on crime. Seasonal effects
could be particularly important for the street crimes under analysis here,
because people spend more time outside when the weather is warmer. Secondly,
the analysis controls for preexisting temporal trends at each camera location.
Failing to control for these pre-camera implementation trends could result in
under or overestimating the cameras’ effects on crime patterns. Examples of
such pre-camera implementation trends include the possibility of regeneration
taking place near a camera location potentially resulting in a generally declining
crime trend—an additional effect beyond simple seasonal variation.
Specific HLM variables are as follows. The length of month variable repre-
sents the number of days per month, given that it is reasonable to expect that
longer months will have higher crime counts. The temporal trend variable
represents the sequential position of the month in the data series in the level-1
HLM equation. This variable captures the linear trend of crime around the
cameras over time at each location. This variable can be positive if crime is
generally increasing, negative if crime is decreasing or zero if the crime trend is
showing no change over time. The ongoing effects of changes over time are
tailored for each location in that each camera location is allowed to have its
own unique linear crime trend over time.
A seasonal effect variable controls for seasonality through the use of a value
to represent the monthly average of the average daily temperature. These
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

756 RATCLIFFE ET AL.
figures were obtained from the historical archives provided by Weather
Underground (available at www.wunderground.com/history). The camera vari-
able represents the effects of camera implementation (0 = non-camera month;
1 = camera implemented). Because of the other variables included, this variable
represents camera implementation while controlling for length of the month,
pre-existing crime trends, and seasonal effects.
At level-1, the units are repeated measurements (monthly observations from
January 2005 through August 2007) on the dependent (crime count) and inde-
pendent (camera implementation, length of month, pre-existing crime trends,
and seasonal effects) variables. These level-1 repeated measures are nested
within the level-2 units (cameras). In this analysis, three dependent variables
are utilized; serious crimes, disorder crimes, and all crime (the sum of both
serious offenses and disorder crimes).
These separate dependent variables were modeled using the same specifica-
tions for the independent variables. All three dependent variables are non-
normally distributed. For this reason, the HLM models are specified as a Poisson
distribution with over-dispersion.
2
Thus, all models follow the specification:
Where Crime Count
it
is the number of crimes occurring within the camera
buffer area for camera i at time t; β
0i
is the mean crime count in camera i in
April 2006 (the mid-point of the study period) when length of month and
seasonal effects are set to their mean for camera i; β
1i
is the slope coefficient
for length of month for camera i; β
2i
is the slope for the linear temporal trends
at camera i; β
3i
is the slope coefficient for the impact of seasonal trends at
camera i; β
4i
is the slope coefficient for the dummy variable representing
camera implementation at camera location i; and r
it
is the residual or unex-
plained variance.
The level-2 model was specified as:
where γ
00
is the average intercept (mean crime count) in April 2006 across all
cameras; γ
10
represents the fixed slope of the length of the month; γ
20
repre-
2. All three dependent variables demonstrate over-dispersion with a standard deviation greater
than the mean: serious crime m = 3.86, sd = 4.48; disorder crime m = 15.60, sd = 18.22; all
crime m = 19.46, sd = 21.51.
Crime Count Length of month Temporal trend
Seasonal effect Camera r
it i i i
iiit
=+ +
+++
ββ β
ββ
01 2
34
()()
()()
βγ
βγ
βγ
βγ
βγ
0000
110
2202
330
440
ii
i
ii
i
i
u
u
=+
=
=+
=
=
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 757
sents the varying slope of ongoing temporal trends; γ
30
represents the fixed
slope of seasonal effects; γ
40
represents the fixed slope of camera implementa-
tion; and u
0i
and u
2i
are the residuals or unexplained between camera variance
in the intercept and temporal trend slope coefficients, respectively.
Weighted Displacement Quotient (WDQ)
Bowers and Johnson’s (2003) WDQ is employed to determine if differences
between the target and buffer areas are a result of displacement from the
target area or a diffusion of benefits from the use of CCTV surveillance in the
target area. The determination of a WDQ first requires the researcher to deter-
mine three operational areas; the target area where the crime reduction strat-
egy has been deployed (in this case, CCTV camera viewsheds), a buffer area
that is estimated to be the most likely location that crime would be displaced
to, and a control area that acts as a check on general crime trends that are
affecting the region in general. The equation for the WDQ is as follows:
where A is the count of crime events in the target area, B is the count of
crime events in the buffer area, C is the count of crime events in the control
area, t
1
is the time since the camera(s) have been active, and t
0
is the pre-
intervention time period (in this case, an equivalent number of months
immediately prior to the installation of the cameras). The examination of the
difference between the buffer and control areas from the pre-intervention to
the intervention period provides the measure of displacement or diffusion into
the buffer area, while the differences between the target and control area
ratios at both times provide the measure of success for the intervention. The
equation above is therefore comprised of both a buffer displacement measure
(Bt
1
/Ct
1
– Bt
0
/Ct
0
) and a success measure (At
1
/Ct
1
– At
0
/Ct
0
).
If the success measure is a positive value, this indicates that the camera
implementation was not successful in reducing crime when compared to the
control area. In this situation—indicating the camera implementation was unsuc-
cessful in reducing crime—then neither the displacement measure nor the WDQ
values are calculated. Only if the success measure indicates a reduction in crime
in the target area is a displacement measure calculated. If the displacement
measure is a positive number, this indicates that when the cameras were imple-
mented, crime went up in the buffer area to a greater extent than in the control
area. This is suggestive of a displacement effect. A negative displacement value
suggests a diffusion of benefits from the target area to the buffer area.
Finally, these two values are combined to form the WDQ value. According to
Bowers and Johnson (2003), WDQ values greater than 1 indicate crime reduc-
tions in the target area and substantial diffusion of benefits to the surrounding
WDQ (B C B C A C A C=− −
tt t t tt t t
11 0 0 11 0 0
//)/(//),
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

758 RATCLIFFE ET AL.
buffer. WDQ values between 0 and 1 indicate diffusion of benefits that are less
than the direct crime reduction effect found in the target area. WDQ values
between 0 and –1 indicate slight displacement from the target area into the
buffer area. WDQ values near –1 indicate displacement effects that offset the
reduction effects seen in the target area. A value near –1 indicates no net
benefit for the program. WDQ values less than –1 indicate displacement effects
much greater than the crime reduction effects in the target area.
Results
The following section will present the results from both the HLM analysis and
the WDQ analysis grouped by crime type: serious crime, disorder crime, and all
crimes. Each crime type was analyzed using both HLM and WDQ.
Serious crime
Results from the initial ANOVA
3
analysis found that there are, on average, 2.67
serious crimes per month per location with significant variation between loca-
tions (p < 0.001). The temporal trend variable showed that there were, on aver-
age, no significant linear crime trends during the time period under analysis
here. Across all camera locations, crime was neither rising nor falling during the
time period (as reflected by the temporal trends variable). The results also
found no evidence of significant seasonal trends. The HLM analysis of serious
crimes found that camera implementation had no significant impact upon the
amount of crime in the target area. Serious crime decreased slightly after
camera implementation, by about 5%, but this drop was not statistically signifi-
cant. Overall, the inclusion of days per month, temporal trends, seasonal
effects, and camera implementation explained about 3.4% of the variance in
serious crime count. Table 2 presents the results from the HLM analysis using
serious crime as the dependent variable.
Table 3 shows each camera site, followed by the percentage change (post-
camera implementation compared to an equivalent period pre-implementation)
in crime level in the target area, the buffer area and the control area for that
site. If these values are positive numbers, then this indicates that crime
increased in the area during the time when the camera was functioning at that
location, compared to an equivalent number of months before the introduction
of the cameras. For serious crimes, four camera locations (sites 3–5, 6, 9, and
10) demonstrated reductions in crime in the target area when compared to the
control area. For these locations displacement measures and WDQ values were
calculated. The WDQ values indicated that sites 3–5, 6, and 10 showed crime
reduction in the target area and a diffusion of benefits into the surrounding
3. ANOVA analyses were conducted with no predictor variables entered into the HLM equation.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 759
buffer area. Site 9 shows crime reduction in the target area and more modest
crime reduction in the surrounding buffer area. Table 3 presents the results of
the WDQ analysis for serious crime.
Disorder crime
Results from the ANOVA analysis finds that there are, on average, 8.34 disorder
crimes per month per location with significant variation between locations
(p < 0.001). Table 4 provides the results of the HLM analysis when investigating
the impact of cameras upon disorder crimes. As expected the length of month
was a significant predictor of higher crime counts. The temporal trend variable
showed that the number of disorder crimes increased slightly during the study
period, on average, across all locations. The expected count of disorder crimes
increased, on average, about 1.3% every month across the evaluation sites. The
significance of the seasonality variable indicates that there are more disorder
crimes in warmer months. The coefficient for the camera variable indicates
that camera implementation significantly reduced disorder crime in the target
area. After camera implementation, the average expected disorder crime count
for the target areas was 16% lower, after controlling for all other variables.
Overall, the inclusion of days per month, temporal trends, seasonal effects, and
camera implementation explained about 10.4% of the variance in disorder crime
count.
Table 2 HLM results for serious crime
1
Fixed Effects Coefficient (S.E.) Event rate ratio Confidence interval
Number of days per
month
0.064
(0.034)
1.066 0.997–1.140
Temporal trend 0.002
(0.006)
1.002 0.989–1.015
Seasonality 0.003
(0.002)
1.003 1.000–1.007
Camera −0.050
(0.104)
0.951 0.774–1.169
Random effect Variance component Chi-square Df
Between camera (level-2)
Mean crime count 0.859** 1855.120 9
Temporal trend 0.007* 17.394 9
Within-camera (level-1)
Residual variation 0.889
Total variance explained 3.357%
*p < 0.05, **p < 0.001.
1
Dependent variable specified as Poisson distribution with over-dispersion. Number of days per
month, seasonality, and camera implementation dummy were specified as fixed slopes. The
temporal trend variable was specified as varying slope.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

760 RATCLIFFE ET AL.
Table 3 Weighted displacement quotient (WDQ) for serious crimes
1
Site
Camera
Type Target Buffer Control
Success
measure
Displacement
measure WDQ Interpretation of WDQ
1 PODSS 0.0 17.0 −0.8 0.00004 Did not reduce crime in the target area
2 PODSS 14.3 6.6 6.8 0.00040 Did not reduce crime in the target area
3–5 PODSS −28.9 −20.5 −2.4 −0.00163 −0.003 1.852 Camera reduced crime, and there was strong diffusion of benefits
6 PODSS −18.2 −15.0 −10.9 −0.00021 −0.001 2.914 Camera reduced crime, and there was strong diffusion of benefits
7 PTZ 11.1 −7.7 1.3 0.00279 Did not reduce crime in the target area
8 PTZ 14.6 7.1 −0.3 0.00398 Did not reduce crime in the target area
9 PTZ −15.0 5.4 9.4 −0.00583 −0.002 0.350 Camera reduced crime, and there was some diffusion of benefits
10 PTZ −21.4 −9.2 5.4 −0.00540 −0.010 1.832 Camera reduced crime, and there was strong diffusion of benefits
1
Target, buffer, and control categories are expressed as percent change between the number of crimes pre- and post-camera implementation. Displacement
measure and WDQ are not calculated if the success measure does not indicate crime reduction in the target area.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 761
WDQs were calculated using disorder crimes as the outcome variable. Sites 3–
5, 8, 9, and 10 had success measures indicating a reduction in crime after the
implementation of the cameras; therefore, displacement measures were
calculated for these locations. For these four sites, only one had a negative
displacement measure. Overall, sites 1, 2, 6, and 7 had no reduction in crime
after implementation of the cameras. Sites 3–5 and 10 had reductions in the
target area but these reductions were offset by displacement into the surround-
ing areas. Site 8 had a reduction in crime in the target area but slight increases
in the buffer area, though there was an overall reduction in crime. Only camera
9 demonstrated a crime reduction in the target area with a diffusion of benefit
into the buffer area. Table 5 presents the findings from the WDQ analysis for
disorder crimes.
All Crime
The final analysis combined serious crime with disorder events. This had the
effect of weighting each incident equally, and is reported here as a measure of
the value of CCTV cameras for all incidents, irrespective of seriousness. Table 6
provides the results of the HLM analysis when utilizing all (both serious and
disorder) crime. Results from the ANOVA analysis find that there are, on aver-
age, 11.35 crimes per month per location with significant variation between
locations (p < 0.001). The temporal trend variable was not significant; there was
no change in trend during the study period. The length of month variable was
significant with each additional day being related to a 5% increase in the
expected count per camera. Crime counts were also significantly higher during
months with higher temperatures. The implementation of cameras significantly
Table 4 HLM results for disorder crime
1
Fixed effects Coefficient (S.E.) Event rate ratio Confidence interval
Number of days per month 0.052* (0.023) 1.054 1.007–1.103
Temporal trend 0.013* (0.006) 1.013 1.001–1.026
Seasonality 0.006** (0.070) 1.006 1.004–1.009
Camera −0.174* (0.001) 0.840 0.733–0.963
Random effect Variance component Chi-square Df
Between camera (level-2)
Mean crime count 1.578** 6357.731 9
Temporal trend 0.000** 68.442 9
Within-camera (level-1)
Residual variation 0.1627
Total variance explained 10.421%
*p < 0.05, **p < 0.001.
1
Dependent variable specified as Poisson distribution with over-dispersion. Number of days per
month, seasonality, and camera implementation dummy were specified as fixed slopes. The
temporal trend variable was specified as varying slope.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

762 RATCLIFFE ET AL.
Table 5 Weighted displacement quotient (WDQ) for disorder crimes
1
Site
Camera
type Target Buffer Control
Success
measure
Displacement
measure WDQ Interpretation of WDQ
1 PODSS 19.2 −20.2 −9.8 0.0023 Did not reduce crime in the target area
2 PODSS 11.8 −8.8 −6.7 0.0017 Did not reduce crime in the target area
3–5 PODSS −1.1 9.4 3.8 −0.0002 0.001 −3.454 Camera reduced crime, but displacement negated gains
6 PODSS −3.0 −26.5 −16.7 0.0003 Did not reduce crime in the target area
7 PTZ 0.9 0.0 −4.1 0.0008 Did not reduce crime in the target area
8 PTZ −16.4 10.9 5.4 −0.0072 0.001 −0.209 Camera reduced crime, but there was slight displacement (net gain)
9 PTZ −35.9 −21.9 −6.7 −0.0107 −0.005 0.479 Camera reduced crime, and there was some diffusion of benefits
10 PTZ −2.7 21.0 8.1 −0.0040 0.009 −2.267 Camera reduced crime, but displacement negated gains
1
Target, buffer, and control categories are expressed as percent change between the number of crimes pre- and post-camera implementation. Displacement
measure and WDQ are not calculated if the success measure does not indicate crime reduction in the target area.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 763
reduced the number of crime events within the target areas. The months
following the implementation of the cameras saw a statistically significant
13.3% reduction in expected crime counts after controlling for the other factors
(p < 0.05). Overall, the inclusion of days per month, temporal trends, seasonal
effects, and camera implementation explained about 12.9% of the variance in
the crime count.
WDQ values were calculated for all crimes. Camera locations 3–5, 8, 9, and
10 demonstrated crime reductions in the target area after implementation of
the cameras. Sites 3–5 and 9 saw some diffusion of benefits into the surrounding
buffer area. Site 8 saw a slight displacement of crime into the buffer area,
though the crime reduction in the target area was enough to offset the
effects of displacement. Finally, at site 10, the reduction of crime in the target
area was offset by the displacement of crime into the surrounding buffer area.
Table 7 reports the results from the WDQ analysis on all crimes.
Discussion
Overall, results from the HLM analysis suggest that the introduction of the
cameras was associated with a 13% reduction in all crime in the target areas
surrounding CCTV implementation sites. The reduction was statistically
significant, after controlling for general temporal trends at each camera site,
seasonality, and the number of days in each month. This reduction was largely
Table 6 HLM results for all crime
1
Fixed effects Coefficient (S.E.) Event rate ratio Confidence interval
Number of days per month 0.055*
(0.020)
1.056 1.014–1.099
Temporal trend 0.010
(0.005)
1.010 0.999–1.020
Seasonality 0.006**
(0.001)
1.006 1.003–1.008
Camera −0.142*
(0.061)
0.867 0.768–0.979
Random effect Variance component Chi-square Df
Between camera (level-2)
Mean crime count 1.292** 6843.160 9
Temporal trend 0.000** 49.689 9
Within-camera (level-1)
Residual variation 1.590
Total variance explained 12.855%
*p< 0.05, **p < 0.001.
1
Dependent variable specified as Poisson distribution with over-dispersion. Number of days per
month, seasonality, and camera implementation dummy were specified as fixed slopes. The
temporal trend variable was specified as varying slope.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

764 RATCLIFFE ET AL.
Table 7 Weighted displacement quotient (WDQ) for all crimes
1
Site
Camera
Type Target Buffer Control
Success
measure
Displacement
measure WDQ Interpretation of WDQ
1 PODSS 14.3 −12.3 −6.9 0.0015 Did not reduce crime in the target area
2 PODSS 12.2 −5.4 −3.8 0.0013 Did not reduce crime in the target area
3–5 PODSS −10.6 0.0 2.2 −0.0006 −0.0003 0.508 Camera reduced crime, and there was some diffusion of benefits
6 PODSS −6.8 −24.3 −15.6 0.0002 Did not reduce crime in the target area
7 PTZ 4.7 −2.3 −2.9 0.0014 Did not reduce crime in the target area
8 PTZ −9.1 9.6 3.7 −0.0040 0.0017 −0.433 Camera reduced crime, but there was some displacement (net gain)
9 PTZ −34.0 −17.0 −4.9 −0.0102 −0.0043 0.424 Camera reduced crime, and there was some diffusion of benefits
10 PTZ −5.4 14.2 7.5 −0.0042 0.0047 −1.114 Camera reduced crime, but displacement negated gains
1
Target, buffer, and control categories are expressed as percent change between the number of crimes pre- and post-camera implementation. Displacement
measure and WDQ are not calculated if the success measure does not indicate crime reduction in the target area.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 765
due to a decline in disorder offenses, as the frequency of serious crimes around
each camera location was generally too low to detect a measurable impact in
serious crime alone.
This does not mean that serious crime was not impacted. It is worth noting
that one would normally expect seasonality and the length of month to be signif-
icant in an analysis of serious street crime. Because the coefficients for these
variables were not statistically significant, this suggests that a possible cause for
this lack-of-finding is that in each month there were insufficient crimes in the
target area for the technique to detect a statistically significant change. For
example, serious crime in the target area for site 1 was two per month prior to
and after camera installation. The values for the target area of site 2 were even
lower. With this in mind, the lack of statistical significance in the serious crime
category could probably be better interpreted as resulting from a lack of
reported serious crime generally, rather than a failure of the CCTV initiative. In
the same vein, Mazerolle et al. (2002) could not attempt an evaluation of violent
crimes around CCTV locations because of the low base rate in the target areas.
A further cause of the low number of serious offenses within the purview of
each camera may be the manner in which the dependent variable was
constructed. First, rather than including a broad measure of all serious crime
(e.g., Squires, 2003), only those serious offenses that the CCTV cameras were
expected to influence were included in the dependent variable count. Secondly,
rather than setting an arbitrary buffer distance out from each camera location,
individual viewsheds were examined for the PTZ cameras. Had we taken the
approach of Sarno, Hough, and Bulos (1999) and established a fixed buffer of
200 meters (just over 650 feet), we would most likely have included a number
of offenses that occurred out of the view of a camera.
The introduction of CCTV was associated with considerably different impacts
on crime at each site. At half of the sites, crime did not reduce in the target
area. At four sites, serious crime was reduced and there was evidence of a
diffusion of positive benefits to surrounding streets. At some sites, crime was
reduced in the target area but there was apparent displacement to surrounding
streets. Therefore the 13% reduction in overall crime was comprised of very
different behaviors at CCTV evaluation sites.
The results of the WDQ are further complicated by the differences seen
across camera type. Both PTZ cameras and PODSS cameras have examples of
successful crime reduction in their target areas. From a policy perspective
choosing between PTZ cameras and PODSS cameras can be important. These
cameras differ on both their usefulness to police investigations and their initial
installation and ongoing maintenance cost. However, from an outcome perspec-
tive, it may be that the actual camera mechanism may be less important than
the choice of location. This might help to explain why both camera types
appeared to have successes and failures. It may have been the location selected
for camera implementation, not the camera type, that ultimately determined
the crime suppression effects. Of course, this is merely raised as a hypothetical
explanation given the limited number of cameras and sites under examination in
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

766 RATCLIFFE ET AL.
this study. It is certainly a finding from the study that merits further investiga-
tion with a larger study.
Inevitably in a study of this nature there are some limitations. Data were
generously provided by the police department of the City of Philadelphia, and
were provided already geocoded and with most address-specific information
removed (beyond the location coordinates). As such, ground-truthing the
accuracy of the geocoding process employed by the city was not possible. Our
extensive previous experience with PPD data, as well as personal contact with
the GIS Mapping Unit, lead us to have confidence in the accuracy and precision
of their geocoding processes such that we could attempt this study, however, it
should be noted that we did not perform the geocoding ourselves.
Perhaps the biggest limitation of the HLM analysis is its inability to disaggre-
gate the effectiveness of each camera type. An attempt to control for the type
of camera at each location is met with difficulty primarily because there are so
few cameras to analyze. Disaggregating the analysis by the type of camera
leaves too few cases for a robust statistical analysis. This line of inquiry,
however, is a worthy area of further investigation. From a practical perspective,
PTZ cameras and PODSS cameras have substantial differences in both initial and
ongoing monitoring cost. Therefore, in order to better understand the effects of
different cameras at different locations, a WDQ analysis was utilized.
Unlike the HLM analysis, WDQ is not able to incorporate sensitivity to season-
ality patterns or to control for subtle trends in changing crime patterns over
time. It compensates for this by incorporating a control area measurement,
used to adjust the result for differences in an area not related to the target or
displacement zones of the cameras. In other words, the control area provides an
indication of what was happening in unaffected areas, and is a broad indication
of trends over the same period of time as the CCTV intervention. WDQ does,
however, provide the opportunity to measure a general indication of the success
of each evaluation site, something not possible with the HLM analysis.
Policy-makers looking to this study to provide a definitive answer are likely to
be a little disappointed in the ambiguity of the results. The reduction in overall
crime of 13% is welcome, though the WDQ results that suggest some camera
sites were unsuccessful in reducing crime at their locations casts a cloud over
any suggestion of there being a benefit to blanket citywide CCTV coverage. The
finding does fit into the broad pattern of results typified by the multi-site
study by Gill and Spriggs (2005), the meta-analysis of Welsh and Farrington
(2007, p. 46) who found a ‘small but significant desirable effect on crime,’ and
the review by Ratcliffe (2006, p. 19) that found ‘achieving statistically
significant reductions in crime can be difficult.’
Evaluating the effectiveness of CCTV is often confounded by the conduct of
other crime prevention initiatives at the same time, making it difficult to tease
out the benefit of the cameras alone; however, some reviews have noted the
benefit of bundling CCTV with other crime prevention programs as a package of
measures. This may help to overcome the public perception that CCTV is a
‘silver bullet’ that will reduce crime in the absence of any other socio-economic
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 767
or opportunity-related changes to the local environment. Farrington, Gill,
Waples, & Argomaniz (2007) concluded that CCTV could reduce crime in car
parks, but was generally ineffective in residential areas or city centers; they
also suggested that the benefits of CCTV may increase when implemented
alongside improved lighting.
Camera success as an investigative tool (an additional factor beyond crime
reduction worth considering, according to Gill & Spriggs, 2005) is potentially
tied to factors such as operator familiarity with the area under examination, the
likely offenders in an area, the nature of businesses and individuals in a camera
location, and the type of crime common to the camera area. Further potential
factors can also include the ability of local police to have sufficient resources to
respond quickly to any incident viewed on the camera, as well as the nature of
the local geography at a camera location; even the quickest police work can be
hampered by easy and accessible escape routes for offenders. Future studies of
camera effectiveness might care to consider these factors as controls on camera
efficiency in the crime prevention arena.
Finally, we did not examine the issue of public perceptions of safety. It may
be that CCTV is politically palatable if public surveys and interviews indicated
improvements in perception of safety and quality of life within the range of
CCTV even in the face of a lack of crime reduction benefits. The evidence from
the British studies is not optimistic (see Gill & Spriggs, 2005); however, it should
be considered an avenue for further work in the evaluation of CCTV in the USA.
Conclusion
This evaluation finds that when serious and disorder offenses were considered
together crime was reduced by 13% after the implementation of the CCTV
cameras while controlling for length of month, seasonal effects, and the unique
temporal trends at each camera. WDQs were then utilized in an attempt to
disaggregate this finding by location and camera type. While there was evidence
that camera implementation had positive effects, the fact that crime did not
reduce in the surveillance areas of half the sites examined cannot be ignored.
Given the low volume of serious crime at each site (as measured on a monthly
basis), it may be prudent to prioritize future CCTV sites based on an objective
measure of the volume of crime at each intersection. Furthermore, given that
the PTZ cameras are able to view activity at more than one street intersection,
selection of future sites would be improved by attempting to find clusters of
street intersections and blocks that have crime problems rather than single
corners. If multiple locations can be viewed effectively from a single camera,
this may be a more cost beneficial use of CCTV technology. Finally, the paucity
of CCTV evaluations is a hindrance in advancing our understanding of any crime
prevention benefits of surveillance technology, a gap in the research literature
that criminologists should seek to remedy as soon as possible. Research or not,
cities are moving ahead with CCTV systems, and if criminologists are to remain
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

768 RATCLIFFE ET AL.
relevant in this crime prevention expansion, more experimental or high quality
quasi-experimental research is necessary.
Acknowledgments
The researchers would like to thank Commissioner Charles Ramsey and Deputy
Commissioner Jack Gaittens, Philadelphia Police Department, for supporting
this project and provision of the necessary data. The views expressed herein are
those of the authors and do not necessarily reflect the views or opinions of
Temple University, the City of Philadelphia, or the Philadelphia Police
Department.
References
Bowers, K. J., & Johnson, S. D. (2003). Measuring the geographical displacement and
diffusion of benefit effects of crime prevention activity. Journal of Quantitative
Criminology, 19(3), 275–301.
Brown, B. (1995). CCTV in town centres: Three case studies (Crime Detection and
Prevention Series, Paper 68). London: Home Office.
Bulwa, D. (2008, February 7). New criminal justice chief wants cops monitoring cameras.
The San Francisco Chronicle, p. B–3.
Clarke, R. V. (2004). Technology, criminology and crime science. European Journal on
Criminal Policy and Research, 10(1), 55–63.
Clarke, R. V., & Felson, M. (1993). Introduction: Criminology, routine activity, and ratio-
nal choice. In R. V. Clarke & M. Felson (Eds.), Routine activity and rational choice
(Vol. 5, pp. 259–294). New Brunswick, NJ: Transaction.
Clarke, R. V., & Weisburd, D. (1994). Diffusion of crime control benefits. In R. V. Clarke
(Ed.), Crime prevention studies (Vol. 2, pp. 165–183). Monsey, NY: Criminal Justice
Press.
Cornish, D., & Clarke, R. (1986). The reasoning criminal: Rational choice perspectives on
offending. New York: Springer-Verlag.
Eck, J. (1997). Preventing crime at places. In L. W. Sherman, D. Gottfredson, D. MacKen-
zie, J. Eck, P. Reuter, & S. Bushway (Eds.), Preventing crime: What works, what
doesn’t, what’s promising. Washington DC: National Institute of Justice.
Farrington, D. P., Gill, M., Waples, S. J., & Argomaniz, J. (2007). The effects of closed-
circuit television on crime: Meta-analysis of an English national quasi-experimental
multi-site evaluation. Journal of Experimental Criminology, 3(1), 21–38.
Gill, M., & Spriggs, A. (2005). Assessing the impact of CCTV. London: Research, Develop-
ment and Statistics Directorate [Home Office]).
Goldstein, H. (2003). On further developing problem-oriented policing: The most critical
need, the major impediments, and a proposal. In J. Knutsson (Ed.), Problem-oriented
policing: From innovation to mainstream (pp. 13–47). Monsey, NY: Criminal Justice
Press.
Harada, Y., Yonezato, S., Suzuki, M., Shimada, T., Era, S., & Saito, T. (2004, November
17–20,). Examining crime prevention effects of CCTV in Japan. Paper presented at the
American Society of Criminology Annual Meeting, Nashville, Tennessee.
Hood, J. (2003). Closed circuit television systems: A failure in risk communication?
Journal of Risk Research, 6(3), 233–251.
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

EVALUATING PHILADELPHIA CCTV 769
Jeffery, C. R., & Zahm, D. L. (1993). Crime prevention through environmental design, oppor-
tunity theory, and rational choice models. In R. V. Clarke & M. Felson (Eds.), Routine
activity and rational choice (Vol. 5, pp. 323–350). New Brunswick, NJ: Transaction.
Klein, A. (2008, February 21). Cameras have cut violence, study says: Skeptics suspect
crime ‘displacement.’ Washington Post, p. 4.
Laird, N. M., & Ware, J. H. (1982). Random-effects models for longitudinal data. Biomet-
rics, 38(4), 963–974.
Mazerolle, L., Hurley, D. C., & Chamlin, M. (2002). Social behavior in public space: An
analysis of behavioral adaptations to CCTV. Security Journal, 15(1), 59–75.
Ratcliffe, J. H. (2004). Geocoding crime and a first estimate of an acceptable minimum
hit rate. International Journal of Geographical Information Science, 18(1), 61–73.
Ratcliffe, J. H. (2006). Video surveillance of public places (Problem-oriented guides for
police, Response guides series no. 4). Washington DC: Center for Problem Oriented
Policing.
Ratcliffe, J. H. (2008). Intelligence-led policing. Cullompton, Devon: Willan.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and
data analysis methods. Thousand Oaks, CA: Sage.
Sarno, C., Hough, M., & Bulos, M. (1999). Developing a picture of CCTV in Southwark town
centres: Final report. London: Criminal Policy Research Unit, South Bank University.
Sivarajasingam, V., Shepherd, J. P., & Matthews, K. (2003). Effect of urban closed circuit
television on assault injury and violence detection. Injury Prevention, 9(4), 312–316.
Snijders, T., & Bosker, R. (1999). Multilevel analysis: An introduction to basic and
advanced multilevel modeling. Thousand Oaks, CA: Sage.
Squires, P. (2003). An independent evaluation of the installation of CCTV cameras for
crime prevention in the Whitehawk estate, Brighton. Brighton: Health and Social
Policy Research Centre.
Welsh, B. C., & Farrington, D. P. (2007). Closed-circuit television surveillance and crime
prevention. Stockholm: Swedish National Council for Crime Prevention.
Appendix A
This table lists the offenses examined in this study and their UCR codes. The
categories are either ‘serious’ (indicating a crime from the FBI UCR Part 1 list)
or ‘disorder’ from the (FBI UCR Part 2 list). Both categories are added together
to create the ‘all crime’ category.
UCR code Crime description Category
111–116 Homicide Serious
211, 231 Rape: stranger Serious
300–305 Robbery: on the highway Serious
306–308 Robbery: purse snatch (force or injury) Serious
388–399 Robbery: of vehicle Serious
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009

770 RATCLIFFE ET AL.
UCR code Crime description Category
411–416, 421–426, 471–476 Aggravated assault Serious
510–517, 520–521, 530–537,
540–541, 591–592
Burglary: residential (including attempts) Serious
550–567, 570–587, 593–594 Burglary: non-residential (including attempts) Serious
610, 620, 630 Theft: pocket picking Serious
611, 621, 631 Theft: purse snatching Serious
614, 618, 624, 628, 634,
638, 640, 641, 642, 643, 649
Theft: from vehicle Serious
720, 722, 724, 726, 728 Vehicle theft (including attempts) Serious
710, 721, 723, 727, 730,
741, 743, 725
Recovery of stolen vehicle Serious
801, 802, 813 Simple assault Disorder
807, 817 Resisting arrest Disorder
1402, 1403, 1404, 1405 Vandalism: public Disorder
1406, 1407, 1408, 1409 Vandalism: private Disorder
1420, 1421, 1422, 1423 Graffiti Disorder
1501–1507, 1516–1518 Violation of the uniform firearms act (VUFA):
adult
Disorder
1519 Prohibited offensive weapon: adult Disorder
1531–1534, 1541–1544 Violation of the uniform firearms act (VUFA):
juvenile
Disorder
1535, 1545 Prohibited offensive weapon: juvenile Disorder
1601 Pandering Disorder
1602 Solicitation Disorder
1708 Public indecency Disorder
1710 Statutory sexual assault Disorder
1711 Open lewdness Disorder
1713 Aggravated indecent assault Disorder
1716 Luring Disorder
1801–1807 Drug sales Disorder
1811–1817 Drug mfg., delivery, or possession with intent
to deliver
Disorder
1821–1827 Drug possession Disorder
1907 Gambling on highway Disorder
2404 Disorderly conduct Disorder
2501, 2502 Loitering Disorder
3302 Minor disturbance Disorder
3306 Disorderly crowd Disorder
Downloaded By: [Ratcliffe, Jerry] At: 12:55 14 October 2009