Computing Crime: Information Technology, Police
Effectiveness, and the Organization of Policing∗
University of Chicago
University of Chicago
December 4, 2006
How does information technology (IT) affect the organization of police work?
How does it in turn affect police crime-fighting effectiveness? To answer these
questions, we construct a new panel data set of police departments covering
1987-2003. We find that while IT adoption had substantial effects on a wide
range of police organizational practices, it had, by itself, a negligible impact on
crime-fighting effectiveness. These results are robust to various methods for con-
trolling for agency-level characteristics and the endogeneity of IT use. We then
suggest and test two explanations for this puzzle. First, we demonstrate that
use of a particular technology, computerized record-keeping, increased recorded
crime rates. Second, we provide evidence that IT investments only had a sub-
stantial impact on crime clearance rates and crime rates when undertaken as
part of a broad set of complementary organizational practices such as those in
the Compstat program.
JEL Classification: L23, M5, O33, K42
∗Garicano thanks the Toulouse Network for Information Technology (TNIT) and Heaton the Na-
tional Consortium on Violence Research (NCOVR) for financial support. We also thank Daron Ace-
moglu, David Autor, Austan Goolsbee, Robert Topel, and Toulouse Network members for their com-
ments. The authors can be reached at email@example.com and firstname.lastname@example.org.
Crime fighting is essentially an information processing task – police agents must use
the information available at the local and aggregate levels to prevent and solve crimes.1
Thus, we should expect large changes in the cost of processing information to have an
important impact on the organization of police work. In this paper we study the impact
of IT on the organization and effectiveness of policing using a newly constructed panel
data set of police agencies covering the period 1987-2003, which we have merged with
FBI local-level crime data.
Our paper contributes to a large literature on the impact of IT on the organization
of work. Despite the growing size of this literature, our knowledge of IT’s impact is still
spotty, in part due to the lack of availability of firm-level data on organizational change
and information technology adoption over time. Some previous studies of the impact
of IT cannot examine organizational changes because they use the industry as the unit
of analysis [e.g. Stiroh (2002), Autor, Katz, and Krueger (1998), and Berman, Bound,
and Griliches (1994)], or because they rely on a cross-section of firms [e.g. Acemoglu,
Aghion, Lelarge, Reenan, and Zilibotti (2006) and Bresnahan, Brynjolfsson, and Hitt
(2002)2]. Others do follow individual firms over time, but either have no data on
information technology adoption [e.g. Rajan and Wulf (2006), Berman, Bound, and
Griliches (1994)], or no information on organizational change [e.g. Brynjolfsson and
Hitt (2003)]. Only a very small number of previous papers provide firm-level evidence
on the evolution of information technology, skill usage, and organizational change;
notably Caroli and Reenen (2001), which reports data for the 1980s in the UK, and from
the early 1990s for France, and Doms, Dunne, and Troske (1997), who study a panel
of manufacturing firms between 1987 and 1992. Like these papers, our paper utilizes
firm-level data on the evolution of skills, organization, and information technology
1Following Arrow (1974), a large literature studies organizations as information processing and
problem solving institutions – e.g. Radner and Zandt (1992), Bolton and Dewatripont (1994), and
2This paper has panel data on IT and inputs, but cross-sectional information for organizational
adoption. However, it is the first to systematically examine non-manufacturing firms
– in our case, public organizations – and the first to study the majority of the firms
within the observed industry. Moreover, it is the only paper to include a long panel
(16 years) covering most of the period of the recent IT revolution. Finally, by merging
in agency-level data on crime rates and arrest levels, we are able to incorporate rich
productivity measures into our analysis.
We start by studying the impact of computerization on productivity and organiza-
tion using a panel of police departments. Our main identification strategy compares,
controlling for city size and other characteristics, the organization and productivity
of departments that adopted more computing technology to that of departments that
adopted less. Consistent with previous research,3we find that IT adoption and skill
are complementary: departments that adopt IT increase police training and introduce
college requirements for new recruits. The evidence suggests that this increase in train-
ing is primarily related to the need to learn to use new devices, rather than IT-induced
enhancement in the training process. Moreover, adopting departments become larger,
increasingly employ special units, and include a larger fraction of support personnel.
In sum, departments become more highly skilled and their organization in many ways
more complex. Despite these changes, we find little evidence that general IT adoption
resulted by itself in an increase in the effectiveness of police work, as manifested both
in clearance rates and in crime rates. We carefully analyze the generality of these re-
sults, and find them robust to alternative samples (by size, by period, early adopters,
growing versus non-growing cities, etc.) and specifications of the IT measure.
Correctly interpreting the underlying causal mechanisms at work is an important
consideration here–our findings indicating that IT promotes organizational change
could reflect reverse causality or omitted variable bias. Given the nature of organi-
zational change, which often involves simultaneous adjustments on a number of dimen-
sions and which may be driven by factors unobserved to researchers, sorting out causal
pathways can be challenging. Using the available data, we attempt to address several
3See, for example, Autor, Katz, and Krueger (1998) and Lehr and Lichtenberg (1999).
alternative explanations for the strong relationship between IT use and our organiza-
tional measures. By including both year and agency fixed effects in our specifications,
we first remove variation that may be due to systematic differences across departments
(such as geography) and well as macroeconomic trends. We also find our results robust
to inclusion of time trends by state or initial level of computerization.
If, as agencies increase in size, their optimal structure involves increasing use of IT
and changes in organizational form, failure to adequately account for agency size could
suggest a spurious effect of IT on organization. In each of our baseline regressions we
flexibly control for the relevant aspect of agency size or workload. As additional checks,
we rerun our regressions first limiting the sample to the largest and smallest agencies
and then including a full set of agency-size decile and year interactions as controls.
The strong positive relationships between IT, worker training, and worker skill persist
in these specifications.
Poorly managed departments may undergo overhauls that affect both IT use and
organizational variables. Using civil litigation cases filed against an agency in 1987 as a
measure of initial department quality, we uncover little evidence suggesting differential
IT adoption by poorly functioning agencies. Alternatively, younger, dynamic cities,
such as Houston or Seattle, may have unobserved characteristics that promote both
IT use and different bureaucratic evolution. Limiting the sample to shrinking cities or
cities with little population change does not alter our conclusions, however.
We also estimate specifications including leads of IT intensity as additional ex-
planatory variables to assess whether exogenous organizational reform could prompt
IT adoption (reverse causality), but obtain little indication of such effects. Another
possibility is that agencies with larger budgets are able to implement both information
technology and superior organizational practices such as increased training. However,
the strong relationship between IT and organization persists when we directly control
for equipment expenditure in our regressions, suggesting that this relationship is not
driven primarily by resource availability.
As a final check, we employ two different instrumental variables (PC availability in
the broader area and body armor use) that attempt to capture variation in the supply of
and demand for IT exogenous to our organization and effectiveness measures. Although
limited, our instrumental variables analysis supports the hypothesis that IT adoption
leads to organizational change. Taken as a whole, our evidence is most consistent
with a causal effect of information technology on organizational structure, with the
large technological changes driving IT adoption in the broader economy contributing
to both computerization and organizational change in the police sector but having little
apparent effect on productivity.
These findings are puzzling: while computers matter organizationally, their effects
do not show up in the productivity numbers.4We propose two possible explanations
for this puzzle, and test them: improvements in crime measurement and complemen-
First, although some information technologies, such as those that identify crime
‘hot-spots’, should improve deterrence, others could actually worsen crime statistics.
For example, if crime reporting is improved, reported crime rates will increase while
clearance rates will drop. Our data contain detailed questions on computer functions,
such as record-keeping, police dispatch, fleet management, etc. We test for heteroge-
nous effects of different technologies by simultaneously entering record-keeping and
deployment measures in our panel regressions. Offense reports increase by 10% when
computers are used for record keeping. Consistent with this hypothesis, such increases
take place for crimes that are more likely to suffer from under-reporting, e.g. larceny,
rather than those which are severe and thus always reported such as homicide. Deploy-
ment technologies, in contrast, are negatively (albeit weakly) associated with offense
Second, we consider the complementarities hypothesis, first advanced formally by
Milgrom and Roberts (1990).5Although IT by itself may have little impact, its impact
4There are precedents in the public sector for large increases in IT that lead to no observable
efficiency gains. Goolsbee and Guryan (2006), for example, find that more access to the Internet by
schools does not measurably increase student achievement.
5In their analysis of modern manufacturing, Milgrom and Roberts (1990) argue that, given the
existence of complementarities among organizational practices, a range of organizational choices may
may be substantial when introduced within the context of an organizational and human
resource system designed to take advantage of it. In the specific context of police
work, the complementarity hypothesis takes one very salient form: Compstat. The
system of practices summarized by this name was initially introduced in the New York
Police Department by Police Commissioner William Bratton under Mayor Rudolph
Giuliani’s leadership and then spread throughout the country. The program aimed
to combine real-time geographic information on crime with strong accountability by
middle managers in the form of daily group meetings, geographic resource allocation,
and data intensive police techniques. The program was widely credited in the press
and by policymakers with playing a substantial role in the recent precipitous drop in
crime experienced by some cities.6
To test the complementarity hypothesis, we study the impact of information tech-
nology when it is adopted together with skilled officers, new problem-solving tech-
niques, extensive use of ‘output’ information in evaluation and deployment of officers,
and a geographic-based structure.7Although the data available for testing this hy-
pothesis are much shorter and more limited (questions on these type of practices were
only introduced in the survey in 1997), they clearly endorse this hypothesis. We find
crime clearance rates were an average of 2.2 percentage points higher in agencies imple-
menting this integrated set of practices. Similarly, crime rates are negatively associated
with Compstat use. Moreover, the individual practices composing Compstat have no
independent ameliorative impact on crime levels or clearance rates.8
We conclude that IT can increase police effectiveness, but that (1) its impact is
have to be altered together for a particular technological advance to improve efficiency. In the presence
of complementarities success is not “a matter of small adjustments, made independently at each of
several margins, but rather involve[s] substantial and closely coordinated changes in a whole range of
the firm’s activities.” (p. 513)
6Some previous research has disputed the claim of a large effect of Compstat; see, for example,
7Our approach is similar to Ichniowski, Shaw, and Prennushi (1997), who study complementarities
among HRM practices and their impact on productivity. IT is not, however, a focus of their study.
8Again, the causal interpretation of this increase must be qualified. If a system of complementary
changes must be undertaken, the fact that some departments choose not to undertake these changes
may reflect some omitted variable, such as the quality of management of the department, in which
case the 2.2% is biased upwards. This problem is common to a large extent to all of the literature on
organizational change (see e.g. Ichniowski and Shaw (Forthcoming)).
obscured by large increases in recorded crime, and (2) the increase in effectiveness only
takes place when IT is introduced in conjunction with certain organizational practices
oriented to take advantage of new data availability.
2 Data Description
The data are drawn from the Law Enforcement Management and Administrative Sta-
tistics (LEMAS) series, a triennial survey of law enforcement agencies in the United
States covering the years 1987-2003.9Although not specifically designed as a longitu-
dinal survey, the broad coverage of the survey makes it possible to identify numerous
agencies at multiple points of time.10The surveys provide rich data on a wide vari-
ety of police operations, including shift scheduling, equipment usage, agency structure
and functions, officer compensation, and administrative policies. To supplement the
LEMAS data, we have matched the surveyed agencies with annual arrest and offense
data from the FBI’s Uniform Crime Reports (UCR) and place-level demographic data
from the Census where possible.
One of the strengths of this data set is that it contains questions on a variety of
different types of IT use and covers a period of enormous IT expansion. Figure 1 plots
aggregate trends in IT use by police agencies. The upper graph details use of different
types of information technologies, including PC’s, mobile data terminals (typically used
by officers to access vehicle, criminal background, or other information while in the
field), and mainframes and servers. In 1987, fewer than 20% of the surveyed agencies
used any computers, but over the next 12 years computer use showed substantial
increases, with PC use growing more rapidly than more specialized technologies. By
the end of the sample over 90% of responding agencies reported IT use. The large
increase in mainframe and server use near the end of the sample is likely attributable
to the increased importance of the Internet in the latter half of the 1990’s.
9The 1996 survey was conducted in 1997, and an additional survey was conducted in 1999.
10All state police agencies and all agencies with 100 or more officers are automatically sampled
with probability sampling for the remaining agencies. In each year roughly 3000 of the approximately
19000 U.S. law enforcement agencies are represented.
Table 9: Complementarities Between IT and Management Practices in Solving Crimes
Clearance Rate For:
Include demographic controls?
Note: This table reports agency-level regressions of the 1997-2003 average clearance rate (arrest/offenses) on
indicators for a Compstat system as well as individual modern police management practices. Each column
entry reports coefficient estimates from a separate regression with inclusion of controls as specified in the
bottom row of the table. Agencies with a Compstat system simultaneously implemented elements of all five of
the listed management practices in more than half of the sample years between 1997-2003. The demographic
controls are the average percent Black, percent Hispanic, per capita income, poverty rate, and log population
of the area covered by the agency over 1997-2003. Heteroskedasticity-robust standard errors are reported in
parenthesis. * denotes significance at the two-tailed 5% level and ** the 1% level.
Table 10: Complementarities Between IT and Management Practices in Deterring Crimes
Offense Rate For:
Include demographic controls?
Note: This table reports agency-level regressions of the 1997-2003 average offense rate (offenses/population) on indi-
cators for a Compstat system as well as individual modern police management practices. Each column entry reportscoefficient estimates from a separate regression with inclusion of controls as specified in the bottom row of the table.
Agencies with a Compstat system simultaneously implemented elements of all five of the listed management practices
in more than half of the sample years between 1997-2003. The controls are the average percent Black, percent His-
panic, per capita income, poverty rate, and log number of agency employees over 1997-2003. Heteroskedasticity-robust
standard errors are reported in parenthesis. * denotes significance at the two-tailed 5% level and ** the 1% level.
Figure 1: Trends in Technology Use By Police Agencies
% of agencies using technology
IT Use By Computer Type
% of agencies using data files
Data Availability by Type
% of agencies with any computers
Ten or fewer employees
Over 100 employees
Between 11 and 100 employees
IT Use By Agency Size
Construction of Management Practices Measures
To examine the role of complementarities in crime reduction, we require separate
agency-level measures of computerization along with relevant modern police man-
agement practices. Following Weisburd, Mastrofski, McNally, Greenspan, and Willis
(2003), we identify five components of a Compstat system: 1) information technology
for crime data collection and analysis 2) use of skilled officers 3) a problem-solving par-
adigm 4) feedback-based evaluation and 5) a geographic-based deployment structure.
We code individual survey items 0-1 (No/Yes) to construct each of the five practice
measures. The constituent survey questions corresponding to each practice measure
1. Information Technology (3)
• Does the department use computers for crime analysis?
• Does the department use computers for crime mapping?
• Does the department use computers for investigation?
• Does the department maintain computerized data on criminal histories?
• Does the department maintain computerized data on crime incidents?
• Does the department maintain computerized data on stolen property?
2. Skilled Officers (1)
• Are more than 6 months (1040 hours) of training provided for new officers?
• Are new officers required to have previous college experience?
3. Problem Solving (1.5)
• Are officers encouraged to use SARA-type problem solving?
• Are collaborative problem solving criteria used in officer evaluations?
• Does the agency engage in problem solving projects with community groups
or government agencies?
4. Feedback-Based Evaluation (1)
• Is citizen survey information collected and provided to patrol officers?
• Is citizen survey information collected and used for allocating resources?
• Is citizen survey information collected and used for prioritizing crime/disorder
• Is citizen survey information collected and used for redistricting patrol ar-
5. Geographic Deployment (1)
• Are officers assigned to geographic areas?
• Are detectives assigned cases based on geographic areas?
The numbers in parenthesis above correspond to the average number of annual survey
questions that must be answered positively in order for an agency to be classified as
employing a particular management practice. For example, an agency which answered
yes to 2 of the problem-solving questions in 1997 and 1999, 1 in 2000, and 3 in 2003
would have an average problem-solving response of 2≥1.5, so its problem-solving prac-
tice indicator would be coded as 1. We consider a department as having a Compstat
system in a given year if it answered yes to at least one of the constituent survey items
for each of the five practices in a given year. Compstat agencies were agencies with
Compstat systems in at least half of the available survey years. Of the 1768 agencies
in our pooled sample, 11.4% used Compstat, 85.6% information technology for crime
analysis, 13.8% high-skill officers, 38.5% problem-solving practices, 63.6% geographic
deployment, and 41.3% feedback-based evaluation.
Table A-1: Alternative Samples and Specifications
Departmental Size and Complexity
Log(Number of employees)
Number of special units
Total written directives
% officers with arrest powers
% field operations staff
% technical support staff
Worker Skill and Training
College requirement for new officers
Hours of training for new officers
Arrests, Offenses, and Officer Injury
Total crime clearance rate
Violent crime clearance rate
Property crime clearance rate
Total offenses rate
Violent offense rate
Property offense rate
Assaults on officers
Note: This table reports robustness checks of the estimated effect of IT use on organizational and arrest outcomes.
Each table entry reports the results of a separate regression. The controls are the same as those reported for
column IV of Tables 2 and 4. Specifications I and II respectively replace the computer index with an indicator
for computerized crime analysis and an indicator for use of any computing technology. Specification III limits
the sample to agencies with available data in 4 or more years. Specification IV omits observations from 1987.
Specification V limits the sample to agencies with a non-zero computer index in their earliest year of reporting.
Each table entry reports a coefficient estimate from a separate regression. Standard errors clustered on agency are
reported in parentheses. * denotes significance at the two-tailed 5% level and ** the 1% level.
Table A-2: Initial Litigation and IT Use
Cases Per 100
10-20 percentile .964
40-50 percentile 2.11
50-60 percentile 2.58
60-70 percentile 3.42
70-80 percentile 4.60
80-90 percentile 6.69
90-100 percentile 15.3
Note: This table reports regressions of the final level of the IT index
on indicators for deciles for the amount of litigation experienced
by an agency in 1987. Each column reports coefficient estimates
from a separate regression. Coefficients are measured relative to
agencies with no reported litigation cases in 1987. Specification
I limits the analysis to agencies for which IT data was available
in 1999 while specification II includes all agencies with at least
one observation on IT after 1987. Both regressions include the
initial level of the computer index as a control and specification II
includes indicators for the final year in which IT data was available
as additional controls. Heteroskedasticity-robust standard errors
are reported in parentheses. * denotes significance at the two-tailed
5% level and ** the 1% level.