Using Official Crime Data

MUST BE 250 words AT LEAST 3 scholarly citations FROM REQUIRED READING ATTACHED in APA format. 

After reading Chapter 3 of the Mosher textbook, “An Arresting Experiment: Domestic Violence Victim and Perceptions” by Miller, and “Controlling a Jail Population by Partially Closing the Front Door” by Baumer and Adams, discuss the following prompts:

• How important is it to use official crime data?
• What are some issues with using official crime data?
• What public policy changes could be made by relying upon the 2 articles?

386

Controlling a Jail
Population by Partially
Closing the Front Door

An Evaluation of a “Summons
in Lieu of Arrest” Policy

Terry L. Baumer
Indiana University Purdue University, Indianapolis
Kenneth Adams
University of Central Florida, Orlando

This study reports on an evaluation of a strategy designed to reduce crowd-
ing of a county jail. The local judiciary sought to reduce the jail population
by ordering local police agencies to issue a summons rather than arrest indi-
viduals accused of seven misdemeanor offenses. The study compares all
cases booked during the first 8 months of the policy with all cases booked
during the same months in the previous year. The results indicate that the policy
was implemented, that it did reduce the intake population, and that there were
minimal side effects; however, the potential impact was considerably overes-
timated in the planning stage.

Keywords: jails; overcrowding; summons; alternatives to arrest

During the past two and one half decades, correctional populations in
the United States have experienced exceptional growth. Between 1980

and 2004, the total number of people under correctional supervision
increased by 280% (Bureau of Justice Statistics, 2005). Although all forms
of corrections experienced increases, the largest changes occurred in the
most restrictive and costly dispositions: prisons and jails. During this same

The Prison Journal
Volume 86 Number 3

September 2006 386-402
© 2006 Sage Publications

10.1177/0032885506291036
http://tpj.sagepub.com

hosted at
http://online.sagepub.com

Authors’ Note: This research was supported by a grant from the Indiana Criminal Justice
Institute. Points of view or opinions are those of the authors and do not necessarily represent
those of the supporting agency. This is a revised version of a paper presented at the annual
meeting of the American Society of Criminology, November 2003.

time frame, prison populations increased 345% and jail populations
increased 288% (Bureau of Justice Statistics, 2005).

These dramatic increases have resulted in crowded conditions for both
prisons and jails. At the end of 2004, state prisons were operating at 99%
of their highest capacity and 115% of their lowest capacity estimates
(Harrison & Beck, 2005b). When the lowest capacity estimate for each
state is used, all but five states exceeded the 90% guideline established by
the American Correctional Association. The situation is similar in local
jails. At midyear 2004, 94% of jail capacity was occupied (Harrison &
Beck, 2005a). The 50 largest jails in the United States hold approximately
31% of the jail population. At midyear 2004, 20 (40%) of these exceeded
their capacity, whereas 33 (66%) were more than 90% full (Harrison &
Beck, 2005a).

At its broadest level, the dynamics of prison and jail populations are the
same. At any given time, the population is a direct function of the number
of admissions and the length of stay (see Cushman, 2002; Pretrial Services
Resource Center, 2000). Although the effect of the former is immediate and
the effect of the latter delayed by the current length of stay, the final result
is the same: Any change to either will result in a corresponding change in
the overall population. In this sense, the sources of the dramatic increases
in prison and jail populations are conceptually the same. A number of
authors have identified policy changes that affected one or both of these
factors for prison populations (Blumstein, 1995; Tonry, 1990).

The factors that drive admissions and length of stay, however, are quite
different for prisons and jails. Much of the prison population is legislatively
driven. In any given jurisdiction the type of sentences (determinate–
indeterminate), type of release (discretionary–mandatory), length of sentence,
extent of credit time, mandatory minimums, sentence enhancements (three
strikes), and a host of other factors are largely controlled by the relevant
sentencing statutes. As a result, significant reductions in prison populations
must rely on statutory changes (or administrative sleight of hand), which
are difficult to come by.

Jail populations, on the other hand, are potentially much more amenable
to change. Nationally, slightly more than 60% of jail inmates are pretrial
detainees (Harrison & Beck, 2005a) who either have been denied bail or do
not have the resources to obtain release through bail. Most, but not all, of
those individuals serving sentences in jail were convicted of misdemeanor or
minor felony offenses. Arrest policies and bail standards are generally estab-
lished at the local level by police agencies and the county courts. Similarly,
misdemeanor sentences seldom suffer the constraints and mandates of their

Baumer, Adams / Summons in Lieu of Arrest 387

felony counterparts. This leaves the nature of the disposition potentially
much more open to negotiations among the interested parties. As a result,
local officials can manipulate both the number of admissions and the length
of stay through changes in local policies (see Cunniff, 2002; Cushman,
2002; Pretrial Services Resource Center, 2000).

This article reports on one approach by a county to control its local jail pop-
ulation. This jurisdiction focused on a “front door” strategy (Blumstein, 1995)
designed to reduce admissions to the county jail system. The executive com-
mittee of the local judiciary ordered police agencies to issue a summons to
appear rather than arrest individuals accused of seven misdemeanor offenses.
At initiation of the policy, it was estimated that this change might reduce
admissions to the county jail system by approximately 20% to 25%. If suc-
cessful, this would have a substantial effect on the local jail population.

Background

Like many others around the United States, the county under study had
a long history of litigation concerning the county jail. In 1972, inmates filed
suit in federal court seeking relief from the overcrowded condition in the
jail. Three years later, in 1975, the judge assigned to the case imposed a cap
on the jail population. The county added capacity to the jail on at least three
separate occasions, but by 1999 the crowding had backed up to include the
county lockup facility. In that year, the population in the county lockup was
added to the existing litigation, and later that year the federal court imposed
a population cap of 213 on the lockup facility. Two years later, with the
mutual assent of the county and the plaintiffs, the cap was raised to 297.

The litigation continued with regular reviews and hearings by the federal
court, but the county was doing little to abate the chronic crowding in the
facility. In April 2002, the federal judge handling the case held county offi-
cials in contempt for their failure to comply with the agreed-on cap of 297
and indicated that financial penalties, and potentially contempt citations,
would be imposed for violations of the cap after May 1. The county was
now on notice that something must be done to control the population of the
county lockup or they would pay the price.

In response to the federal judge’s action, the executive committee of the
county court system, noting “its obligation to assist the Sheriff and other
county officials in complying with the Federal Court Order and to maintain
public safety within our community,” issued a court order on April 18,
2002, designed to help control the population of the county lockup facility.

388 The Prison Journal

This order noted the need to comply with the population cap and, pursuant
to that goal, established a “summons in lieu of arrest” policy for seven non-
violent, misdemeanor offenses: possession of marijuana, possession of
paraphernalia, driving with a suspended license, operating a vehicle never
having received a license, prostitution, patronizing a prostitute, and conver-
sion (generally shoplifting). The order did not apply to individuals charged
with the felony versions of these offenses.

This order contained two substantive provisions. The first ordered the
sheriff to advise all law enforcement agencies operating within the county
to issue a summons (a ticket) in lieu of arrest for these offenses. This
applied to any combination of these seven offenses and any nonarrestable
infraction or ordinance violations that might be included in the same inci-
dent. If the individuals had any other criminal charges or an outstanding
warrant (even for one of the eligible offenses), the sheriff could still accept
and book them into the lockup just like any other criminal offense.

Because the above order was contrary to long-standing policies within
the county and many of its constituent police departments, the executive
committee anticipated a potential for noncompliance: Simply telling the police
agencies within the county to stop arresting individuals for these offenses
probably was not going to be very effective. To ensure compliance with the
new policy, the second provision ordered the sheriff to stop accepting, at the
lockup facility, individuals charged only with the above offenses. Thus, if a
particular department or individual officer arrested an individual for one of
the eligible charges, the sheriff’s department was instructed to turn them
away by refusing to book them into the lockup facility. The court ordered
the sheriff to advise all agencies within the county of this policy.

Although no formal analysis was conducted prior to issuance of the
order, it appears that a substantial impact on the lockup population was
anticipated. The order noted, “During an average week, the lock-up
receives between 180-250 individuals charged with [the above] non-violent
misdemeanor offenses.” No indication was given as to whether the policy
was expected to apply to all of these individuals or some subset of them.
Assuming the former, the anticipated impact on the intake population
would have been a reduction of between 26 and 36 individuals per day.
Given that the county booked approximately 142 people per day (slightly
fewer than 1,000 per week), the policy offered the potential to reduce the
intake population by 18% to 25%.

The impact on the total lockup population would depend on the length
of stay for these individuals. For example, if prior to implementation of the
new policy, the individuals charged with these minor offenses were booked

Baumer, Adams / Summons in Lieu of Arrest 389

out within 24 hours, the impact would be between 26 and 36 people. This
would be a reduction of approximately 10% (10% of 297 = 29.7). However,
if prior to implementation of the policy these individuals stayed 2 days, the
lockup population would be reduced by between 52 and 72 people (20%).
Under any of these scenarios, the projected impact of the court order would
be substantial.

The following analysis will focus on three areas related to the policy.
First, implementation of the policy will be reviewed. An initial analysis will
assess the actual size of the target population as defined by the court order
and interpreted by the sheriff’s department employees at the lockup. These
estimates will form the outer boundaries of potential for the outcome analy-
sis. This will be followed by an analysis of the extent of implementation for
the policy.

Next, the primary impact of the policy on the county lockup facility will
be assessed. Given that the policy was explicitly designed to divert individu-
als charged with the seven misdemeanor offenses from lockup, the reduction
in number of lockup bookings will be investigated. As noted above, the effect
of the policy on the overall lockup population depended on both the extent of
implementation and the length of stay for the target population. The impact
of the policy on length of stay and total “bed days” will be assessed.

Finally, secondary outcomes of the new policy will be reviewed. Although
the court order issued by the county executive committee did not address pos-
sible secondary outcomes for the new policy, a number of plausible hypothe-
ses are possible. For example, it would be reasonable to anticipate an increase
in the failure to appear (FTA) rate for the target cases. In addition, it might be
hypothesized that the new policy would affect case disposition in a number
of ways. The analysis will look at the number of cases without a disposition
at least 10 months later and the nature of the disposition.

Method

The county stored information for all criminal cases on a mainframe
case management system. The researchers worked with a county program-
mer to generate cases from the first 8 months of the new policy period and
a comparison group selected from the same period of the preceding year.

The time frame was dictated by a policy revision made by the county.
From the time of implementation on April 19, 2002, criminal justice offi-
cials were under some pressure to rescind the order. In particular, some
neighborhood groups objected strenuously to the issuance of citations for

390 The Prison Journal

prostitution. They argued that issuing tickets for prostitution did nothing to
reduce prostitution in their areas of the city. During the summer and fall of
2002, the policy became one of the issues in the election for county sheriff,
with the eventually winner calling the county a “laughingstock” for issuing
citations for misdemeanor prostitution. The judges revised the order by
removing prostitution from the list of eligible offenses on December 20,
2002—almost exactly 8 months after the original order. The present study
focused on cases originating during the initial 8-month period when all
seven offenses were included.

Selection of cases was the same for both 2002 and the comparison group
from the previous year. For the primary analytic files, all cases that included
at least 1 of the 7 charges and that fell between April 19 and December 20
were selected. The files included information on all charges associated with
this case (level, type), date of booking, date of disposition, nature of dispo-
sition for all charges, and basic characteristics of the individual charged in
the case (race, sex, date of birth). Because case was the unit of analysis,
individuals might be included multiple times. This generated 6,110 cases
from the target year and 6,221 for the comparison year. Because all cases
occurring in the county during the sampling frame were included and it
cannot be inferred that these cases represent a random sample of cases in
other jurisdictions, no statistical tests of significance are reported.

Results

Eligible Cases and Level of Implementation

The general parameters of the target population are presented in Table 1.
The number of cases with any of the seven misdemeanor offenses declined
slightly from 6,221 for the comparable period of the previous year to 6,110
during the 8-month study period. However, cases covered by the summons in
lieu of arrest order increased from 58.6% to 65.8% of all cases with one or
more of the seven charges. This amounted to an increase of 379 cases in which
individuals were charged with one, or more, of the misdemeanor target
offenses and no other criminal offenses. Overall, 4,022 cases were potentially
eligible for a citation only during the first 8 months of the policy, whereas
3,643 would have been eligible during the same 8 months in the prior year.

The above findings indicate that the potential impact of the change in
policy was considerably lower than suggested in the court order. The original
order noted that the target cases accounted for between 180 and 250 cases per

Baumer, Adams / Summons in Lieu of Arrest 391

392 The Prison Journal

week. When translated to the 8-month study period (243 days), this estimate
would be between 6,245 and 8,675 cases. The total number of cases with at
least one of these charges (6,110) was fairly close to the lesser of the two esti-
mates. Because the total number of cases is similar for each period, the sug-
gestion is that the lower estimate of 180 per week was actually the more
accurate of the two. However, when cases with other criminal charges are
excluded, the number of eligible cases (4,022) was only 65.8% of this esti-
mate during the study period and only 58.6% in the comparable period the
preceding year. This overestimate of the target population limited the poten-
tial impact of the policy change to less than two thirds the original estimate.

Although the target population was smaller than anticipated, with full
implementation the summons in lieu of arrest policy could still substan-
tially reduce the number of people booked into the county lockup. Four
types of booking were possible for the eligible cases: (a) An “outright”
booking occurred when the officer made an arrest and the defendant was
brought to lockup, (b) a “summons” booking occurred when the officer
issued a citation and the defendant was booked when he or she appeared in
court, (c) a “warrant” booking occurred when the defendant was arrested on
a warrant for one of the targeted offenses, and (d) “no booking” occurred
when the defendant was cited by the officer but failed to appear and was
never arrested on the subsequent warrant. Cases subject to the summons in
lieu of arrest policy could be any of the latter three types, although it explic-
itly sought to eliminate outright bookings for the targeted offenses.

Table 2 presents the type of booking for eligible cases. This table shows
that under the summons in lieu of arrest policy, only 20.2% of the eligible
cases experienced outright bookings, whereas for the comparison period,
59.5% were outright bookings.

These numbers have double implications for policy implementation.
First, these figures could be interpreted as an indication of 80% compliance

Table 1
Target Charges and Eligible Cases

Prior to Policy During Policy

n % n %

Eligible cases 3,643 58.6 4,022 65.8
Not eligible cases 2,578 41.4 2,088 34.2
Total 6,221 100.0 6,110 100.0

Baumer, Adams / Summons in Lieu of Arrest 393

with the court order not to arrest these individuals. Consultation with sheriff’s
department personnel who worked in the lockup during this time indicates
that an outright booking for an apparently eligible case could occur in
several ways. If an officer stopped an individual for an eligible offense and
discovered an outstanding warrant for that individual from another case, the
officer was obliged to make an arrest. This resulted in both a warrant book-
ing for the old case and an outright booking for the new offense. Without
the old warrant, the person may have received a citation only. Another sit-
uation occurred when an officer arrested an individual for an offense eligi-
ble for a citation and brought him or her to lockup for booking, and the
booking officers entered the information into the case management system
before noticing that the individual should not have been arrested. Another,
less common situation occurred as above, but the arresting officer had left
the lockup before the processing officers noticed that the case should not be
processed as an outright arrest. A fourth exception occurred when the pro-
cessing officers noted that the police officer had arrested a summons in lieu
of arrest case, but the arresting officer refused to take the defendant back
and issue a citation. In these cases, rather than fight about the correct pro-
cessing of the case, the processing officers tended to go ahead and book it
as an outright case. Thus, the figures in Table 2 clearly indicate substantial
compliance with the court order.

However, Table 2 indicates a second, more serious, complication for the
potential impact of the new policy. A substantial number of cases were han-
dled in a way consistent with the summons in lieu of arrest policy even
before its implementation. During the comparison period, a full year before
implementation, only 59.5% of the eligible cases involved an arrest and
outright booking, whereas 25.4% involved a summons booking. Thus, the

Table 2
Type of Booking for Eligible Cases

Prior to Policy During Policy

Type of Booking n % n %

Outright arrest 2,166 59.5 814 20.2
Warrant 338 9.3 727 18.1
Summons 926 25.4 1,942 48.3
Never booked 213 5.8 539 13.4
Total 3,643 100.0 4,022 100.0

394 The Prison Journal

target population, of people actually arrested for one of the target offenses,
was only about 60% the size of the original estimate. Thus, in addition to
the overestimate of the number of eligible cases noted earlier, the announced
policy represented only an incremental change in existing practices. The
result was that the potential for the policy was about 40% that estimated in
the court order (.658 × .595 = .392). Rather than having the potential of
reducing the intake population by 180 to 250 people per week, the more
realistic figure was 71 per week (about 10 per day).

Primary Outcomes

Lockup bookings. The summons in lieu of arrest policy was intended to
directly reduce the number of bookings at the county lockup. Specifically,
it was directed at a reduction in the number of outright bookings at the facil-
ity. Individuals who were cited for the target offenses would still be booked
when they appeared in court, but this was accomplished on the nonsecure
side of the lockup, which was not part of the federal court order. However,
warrant bookings were processed through the lockup facility just as out-
right bookings. To the extent that the new policy reduced outright bookings
but increased warrant bookings, its impact would be limited. Multiple
bookings for specific cases, usually created by multiple arrests on warrants,
could also limit the policy impact.

Both the number and percentage of outright bookings decreased during the
study period (Table 3). During the comparison period, cases with one or more
of the target offenses accounted for 4,589 outright bookings, or 73.8% of the
cases. During the study period, however, these numbers dropped to 2,634 out-
right bookings (43.1%). The difference between the two periods was 1,955

Table 3
Type of Booking for All Cases With One or More Target Charge

Prior to Policy During Policy

Type of Booking n % n %

Outright 4,589 73.8 2,634 43.1
Warrant 427 6.9 900 14.7
Summons 976 15.7 2,002 32.8
No booking 229 3.7 574 9.4
Total 6,221 100.1 6,110 100.0

Baumer, Adams / Summons in Lieu of Arrest 395

fewer outright bookings. This impact was moderated considerably, however,
by an increase in the number of warrant bookings, which more than doubled
from 427 to 900. The result was that the number of eligible cases booked
through the county lockup (outright and warrant) dropped 29.6%, from 5,016
during the comparison period to 3,534 during the study period. The difference
of 1,482 amounted to an average of 6.1 fewer cases booked per day (1,482 ÷
243 = 6.1). This is considerably lower than the 26 to 36 per day projected by
the court order and closer to the two-fifths figure (39.2%) identified above.

Another potential impact of the new policy might be through the total
number of outright or warrant bookings for each case. Because of FTA and
other violations of court orders, it is possible that the individual charged in
a single case might have multiple arrests and bookings for that case. For the
present study, the researchers captured the type of booking for up to four
bookings for each case. Table 4 presents the number of lockup bookings
(outright or warrant) for the two study periods. The total number of lockup
bookings for all cases with any of the target offenses dropped from 7,720
during the comparison period to 5,443 during the study period. This
decrease of 2,277 fewer lockup bookings for these cases amounted to 9.4
bookings (2,277 ÷ 243) per day.

Not all of the reduction in lockup bookings, however, can be attributed to
the new policy. If the number of bookings for eligible and noneligible cases is
compared, the reduction for policy-eligible cases was reduced by only about
1,219 bookings (3,616 – 2,397) between the two periods. This amounts to only

Table 4
Number of Outright and Warrant Bookings by Eligible Case

Prior to Policy During Policy

Not Eligible Eligible Not Eligible EligibleLockup
Bookings
Per Case

n % n % n % n %

None 60 —a 1,031 — 80 — 2,256 —
One 1,521 37.1 1,908 52.8 1,305 42.8 1,285 53.6
Two 1,160 28.3 946 26.2 912 29.9 718 29.9
Three 735 17.9 486 13.4 477 15.7 282 11.8
Four 688 16.8 276 7.6 352 11.7 112 4.7
Total 4,104 100.1 3,616 100.0 3,046 100.0 2,397 100.0

Note: For this table, the unit is booking (cases multiplied by the number of lockup bookings).
a. No bookings counts as 0.

396 The Prison Journal

about one half (53.5%) of the total reduction noted above. The remainder
(1,058 lockup bookings) can be attributed to a drop of 25.8% in the number of
bookings for noneligible cases with one or more of the eligible offenses.

Lockup population. All things being equal, fewer lockup bookings
should translate into some relief for the lockup population. The following
analysis looks at the median length of stay and total bed days occupied by
this population. Because no time of day was recorded in the data system for
when an individual was booked into the lockup or when they were released,
the analysis will use the less precise measure of day. Thus, if a person is
booked in and booked out on the same day, as would be the case under the
summons in lieu of arrest policy, their length of stay should be zero. To
make the comparisons meaningful, a cutoff date of October 27 of the fol-
lowing year was enforced for both groups. Cases with no jail start date
and/or no jail end date were excluded.

Cases originating during the summons in lieu of arrest period were more
likely to be booked and released on the same day than were cases during
the comparison period (Table 5). For cases eligible for the summons in lieu
of arrest policy, the percentage booked out on the same day jumped from
49.8% to 67.4%. However, the people charged in these cases tended not to
stay very long either before or during the policy period. The mean length of
stay for eligible cases was only 1.8 days before the policy was implemented
and 1.5 days during the policy period. The longer stays were reserved for
other cases, as reflected in the mean stays of 8.4 and 7.3 days for all cases
with one of the target offenses.

Table 5
Length of Stay and Bed Days Consumed

Prior to Policy During Policy

One or More One or More
Target Offenses Eligible Target Offenses Eligible

Booked and released same day
n 2,372 1,693 2,844 2,342
% 39.9 49.8 51.5 67.4

Stay in days
Mdn 1 1 0 0
M 8.4 1.8 7.3 1.5
Total bed days 49,796 6,024 40,168 5,061

Baumer, Adams / Summons in Lieu of Arrest 397

Total bed days in jail were calculated for both groups. For all cases with
one or more of the target offenses, the total number of bed days occupied
changed from 49,796 for the cases originating during the comparison
period to 40,168 for cases originating during the summons in lieu of arrest
period. This amounted to 9,628 fewer bed days. As a percentage of possi-
ble bed days, using the population cap of 297 and the exposure period of
544 days, these cases accounted for about a 6.0% reduction in total bed
days during the study periods.

Unfortunately, the above reduction was largely the result of factors other
than the summons in lieu of arrest policy. The eligible population consumed
6,024 bed days in the comparison period compared with 5,061 during the
summons in lieu of arrest period. A difference of 963 bed days is attribut-
able to the cases potentially eligible for the new policy. This is about 10%
of the difference noted above and amounts to 0.6% of total bed days during
the study periods. As Table 5 shows, the eligible cases tended to be booked
in and out fairly quickly before the new policy, making a significant impact
on the lockup population difficult to achieve (cf. Cunniff, 2002).

Secondary Outcomes

FTA. Two potential secondary outcomes of the summons in lieu of arrest
policy were reviewed: FTA and case disposition. Large differentials in
either of these could affect the viability of the policy independent of the
effect on the lockup population.

FTA was measured by counting the number of FTA entries in the court
record for each case. The number of FTAs for all cases with a target offense
is presented in Table 6. The percentage of cases with no FTA decreased
from 52.4% in the comparison year to 46.7% following implementation of

Table 6
Failure to Appear (FTA) for All Cases

With One or More Target Offenses

Prior to Policy During Policy

Number of FTAs n % n %

No FTAs 3,258 52.4 2,854 46.7
One FTA 1,977 31.8 2,280 37.3
Two or more FTAs 986 15.8 976 16.0
Total 6,221 100.0 6,110 100.0

398 The Prison Journal

the policy. A corresponding increase from 31.8% to 37.3% was recorded in
the percentage of cases with one FTA. However, the percentage of cases
with two or more FTAs was nearly identical: 15.8% versus 16.0%. Overall,
this amounted to a net increase of 293 cases with one or more FTAs.

The target cases for the summons in lieu of arrest policy had a higher
rate of FTA in both the comparison and treatment periods. Table 7 indicates
that the percentage of target cases with one or more FTAs increased from
51.5% to 60.7% when the policy went into effect. The corresponding
figures for cases with one of the seven offenses but also another criminal
offense, which made them ineligible for a simple citation, actually dropped
from 42.2% to 39.0% with one or more FTAs. As with the figures for the
entire sample, for the target group of eligible cases the percentage with two
or more FTAs remained about the same: 17.1% versus 18.0%.

The FTA rate was even higher for eligible cases treated in compliance
with the summons in lieu of arrest policy (no outright booking; Table 8).

Table 7
Failure to Appear (FTA) Rates for Policy Eligible Cases

Prior to Policy During Policy

Not Eligible Eligible Not Eligible Eligible

n % n % n % n %

No FTAs 1,490 57.8 1,768 48.5 1,274 61.0 1,580 39.3
One FTA 722 28.0 1,255 34.4 564 27.0 1,716 42.7
Two or more FTAs 366 14.2 620 17.1 250 12.0 726 18.0
Total 2,578 100.0 3,643 100.0 2,088 100.0 4,022 100.0

Table 8
Failure to Appear (FTA) for Eligible Cases

With No Outright Booking

Prior to Policy During Policy

Number of FTAs n % n %

No FTAs 524 35.5 1,131 35.3
One FTA 630 42.7 1,460 45.5
Two or more FTAs 323 21.9 617 19.2
Total 1,477 100.1 3,208 100.0

Baumer, Adams / Summons in Lieu of Arrest 399

For the cases occurring after implementation of the summons in lieu of
arrest policy and with no outright booking, only 35.3% had no FTA for their
case, whereas 45.5% recorded one and 19.2% recorded two or more. It is
noteworthy that the percentages for the same group from the comparison
period are virtually the same: 35.5%, 42.7%, and 21.9%, respectively.
These figures have several implications. First, for cases handled with a cita-
tion, there will probably be an initial FTA. However, approximately 80% of
the cases experience no more than one FTA. The ultimate disposition of
these cases is discussed below. Second, given the similarity of the results
between the two periods, the high FTA rate could have been anticipated.

Case disposition. To allow meaningful comparisons of case disposition
between the treatment and comparison cases, a cutoff date of October 27 of
the following year was used for both groups. This would allow a minimum
of approximately 10 months for the last cases selected to be disposed. After
this time frame, 75.3% of all cases with one or more of the target charges
during the summons in lieu of arrest period had been disposed, whereas
80.6% had been disposed in this time frame during the comparison period.
In actual numbers, this translated to 1,209 cases in 2001 and 1,510 cases in
2002 that were still unresolved by the end of October the following year.
However, of the cases eligible for summons in lieu of arrest, 766 remained
open for the comparison period, compared with 1,126 for the policy
period—a difference of 360 more open cases after the same period.

Table 9 summarizes the nature of the outcome for cases reaching dispo-
sition during the above described period. Both before and during the imple-
mentation of the summons in lieu of arrest policy, the majority of all cases
with any eligible charge resulted in a dismissal of all charges. This per-
centage was slightly higher during the policy period (52.9%) than during
the comparison period (50.6%). The percentage of cases with at least one
guilty verdict decreased from 48.9% in the comparison period to 46.4%

Table 9
Type of Disposition for All Cases With an Eligible Charge

Prior to Policy During Policy

Type of Disposition n % n %

All dismissed 2,535 50.6 2,422 52.9
All not guilty 25 0.5 33 0.7
Any guilty 2,447 48.9 2,126 46.4
Total 5,007 100.0 4,581 100.0

400 The Prison Journal

during the summons in lieu of arrest period, whereas the percentage of
cases with all charges not guilty remained about the same (0.5% vs. 0.7%).

Discussion and Conclusions

The target population for the policy was considerably smaller than antic-
ipated. The original court order indicated that between 180 and 250 indi-
viduals were charged weekly for the target offenses (26-36 per day). The
total number of cases including any one of the target offenses in either
the comparison or policy implementation periods almost approximated the
lower of these two numbers but was not close to the 215 implied by the
court order.

The court order further restricted applicability of the policy to arrestees
“who are only charged with the following misdemeanor crimes.” Any case
involving any other arrestable offense was excluded, as were individuals
charged by the officer with a felony version of any of the target offenses. In
addition, individuals with outstanding warrants on other charges were
excluded. These restrictions reduced the eligible cases to fewer than two
thirds of all cases involving the target offenses. Taken together, the above
considerations reduced the potential target population from the projected
26 to 36 per day to fewer than 17 per day.

Implementation issues further complicated the picture. The police depart-
ments in the county did comply substantially with the new policy. Of all cases
with the appropriate mix of charges, only approximately 20% involved
arrests and outright bookings during the first 8 months of the policy. This sug-
gested approximately 80% compliance with the court order. Unfortunately,
this was only an incremental change over existing practices. In the compari-
son period, 1 year prior to the study period, 59.5% of the target cases involved
an arrest and outright booking, with the remaining cases handled in a way
consistent with the summons in lieu of arrest policy. This further reduced the
potential of the policy to 60% of the target cases. When combined with the
overestimate of the target population, the potential impact of the new policy
on the lockup population was only about 40% of the lowest original estimate,
or 10 per day rather than the projected 26 per day.

The impact of the policy on the lockup population was measured in three
ways: the number of cases booked into the lockup, total number of book-
ings for eligible cases, and the number of bed days saved by the policy. The
number of cases booked at the lockup (outright or warrant initial booking)
dropped 29.6% between the comparison and study periods. This decrease
of 1,482 cases amounted to 6.1 fewer cases booked at lockup each day.

Total lockup bookings for each case also declined following implemen-
tation of the policy. During the study period, the total number of lockup
bookings for all cases with at least one of the target offenses decreased by
2,277 after the policy was implemented. Unfortunately, because the total
number of bookings for ineligible cases also declined, only about one half
(53.5%) of this decrease was attributable to cases covered by the summons
in lieu of arrest policy.

Holding time at risk constant, the total number of bed days consumed by
these cases also decreased. For all cases involving at least one of the target
offenses, the number of bed days decreased by 9,628 during an exposure
frame of 544 days. However, only about 10% (963 bed days) of this decline
could be attributed to cases eligible for the summons in lieu of arrest policy.
Even when accompanied by an arrest, the eligible cases in the comparison
period obtained release fairly quickly. Further reductions would be very dif-
ficult. As it turns out, most of the reduction in bed days was attributable to
changes in the length of stay for the noneligible cases.

FTA and case disposition were also investigated as possible secondary
outcomes of the summons in lieu of arrest policy. For eligible cases, the
percentage of cases with one or more FTAs increased from 51.5% in the
comparison period to 60.7% for cases initiated during the first 8 months
of the policy. The corresponding figures for cases with one or more of the
target offenses, but additional criminal charges, dropped from 42.2% to
39.0%. This resulted in a net increase of 293 cases with one or more FTAs.

The primary change in case disposition was for the percentage with any
disposition. For both groups, the time available was held constant to
approximately 18 months from initial case selection. During this period, the
percentage of cases with any disposition decreased from 80.6% for the
comparison period to 75.3% for the cases initiated. Eligible cases experi-
enced a similar decrease from 79% disposed to 72% disposed after the
same period. The net number of cases not disposed 18 months after the
beginning of the study period increased by 310 for all cases with one or
more of the target offenses. However, 360 more eligible cases remained
open after comparable time frames.

Successful initiatives require both careful design and full implementa-
tion. In the present case, the idea to control the county jail population
through a reduction in the number of arrests was a viable approach.
However, the target population was overestimated, and many of the cases
were processed in compliance with the new policy even before it was
implemented. Although the program evaluation literature is littered with
examples of programs or policies hampered by partial implementation, this

Baumer, Adams / Summons in Lieu of Arrest 401

was not the problem for this county. The effects of the new summons in lieu
of arrest policy were in the projected direction, but the impact fell consid-
erably short of expectations, primarily because of design and planning fail-
ures. More detailed data analysis and planning could have identified these
issues during the policy formation period.

In the present case, the financial cost of implementation was minimal,
and the substantive outcomes were small, but positive. However, it does not
always turn out this way. Substantially overestimating the size of the target
population or not understanding the exact nature of current practice can, at
best, as was seen in this case, dilute the potential impact of a proposed
change. In other situations, the changes can be both financially and politi-
cally expensive while making minimal improvement in the situation.

References

Blumstein, A. (1995). Prisons. In J. Q. Wilson & J. Petersilia (Eds.), Crime (pp. 387-419). San
Francisco: ICS Press.

Bureau of Justice Statistics. (2005). Number of persons under correctional supervision [Table].
Retrieved March 20, 2006, from http://www.ojp.usdoj.gov/bjs/glance/tables/corr2tab.htm

Cunniff, M. (2002). Jail crowding: Understanding jail population dynamics (NIC 017209).
Washington, DC: U.S. Department of Justice, National Institute of Corrections.

Cushman, R. (2002). Preventing jail crowding: A practical guide (NIC 016720). Washington,
DC: U.S. Department of Justice, National Institute of Corrections.

Harrison, P., & Beck, A. (2005a). Bureau of Justice Statistics bulletin: Prison and jail inmates
at midyear 2004 (NCJ 208801). Washington, DC: U.S. Department of Justice, Office of
Justice Programs.

Harrison, P., & Beck, A. (2005b). Bureau of Justice Statistics bulletin: Prisoners in 2004 (NCJ
210677). Washington, DC: U.S. Department of Justice, Office of Justice Programs.

Pretrial Services Resource Center. (2000). A second look at alleviating jail crowding: A systems
perspective (NCJ 182507). Washington, DC: U.S. Department of Justice, Office of Justice
Programs, Bureau of Justice Assistance.

Tonry, M. (1990). Malign neglect: Race, crime, and punishment in America. New York:
Oxford University Press.

Terry L. Baumer is an associate professor in the School of Public and Environmental Affairs
at Indiana University Purdue University, Indianapolis. He has published work in the areas of
fear of crime, electronic monitoring, and drug dependence in arrestees. He most recently com-
pleted an evaluation of a new Arrestee Processing Center.

Kenneth Adams is a professor of public affairs in the College of Health and Public Affairs at
the University of Central Florida. His areas of expertise include mentally ill offenders, police–
community relations, institutional corrections, and evaluation of crime control strategies, such
as gun control and juvenile curfews.

402 The Prison Journal

<< /ASCII85EncodePages false /AllowTransparency false /AutoPositionEPSFiles true /AutoRotatePages /None /Binding /Left /CalGrayProfile (Dot Gain 20%) /CalRGBProfile (sRGB IEC61966-2.1) /CalCMYKProfile (U.S. Web Coated \050SWOP\051 v2) /sRGBProfile (sRGB IEC61966-2.1) /CannotEmbedFontPolicy /Error /CompatibilityLevel 1.3 /CompressObjects /Off /CompressPages true /ConvertImagesToIndexed true /PassThroughJPEGImages true /CreateJDFFile false /CreateJobTicket false /DefaultRenderingIntent /Default /DetectBlends true /ColorConversionStrategy /LeaveColorUnchanged /DoThumbnails false /EmbedAllFonts true /EmbedJobOptions true /DSCReportingLevel 0 /EmitDSCWarnings false /EndPage -1 /ImageMemory 1048576 /LockDistillerParams true /MaxSubsetPct 100 /Optimize false /OPM 1 /ParseDSCComments true /ParseDSCCommentsForDocInfo true /PreserveCopyPage true /PreserveEPSInfo true /PreserveHalftoneInfo false /PreserveOPIComments false /PreserveOverprintSettings true /StartPage 1 /SubsetFonts true /TransferFunctionInfo /Apply /UCRandBGInfo /Remove /UsePrologue false /ColorSettingsFile () /AlwaysEmbed [ true ] /NeverEmbed [ true ] /AntiAliasColorImages false /DownsampleColorImages true /ColorImageDownsampleType /Bicubic /ColorImageResolution 300 /ColorImageDepth -1 /ColorImageDownsampleThreshold 1.50000 /EncodeColorImages true /ColorImageFilter /DCTEncode /AutoFilterColorImages true /ColorImageAutoFilterStrategy /JPEG /ColorACSImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >>
/ColorImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >>
/JPEG2000ColorACSImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >>
/JPEG2000ColorImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >>
/AntiAliasGrayImages false
/DownsampleGrayImages true
/GrayImageDownsampleType /Bicubic
/GrayImageResolution 300
/GrayImageDepth -1
/GrayImageDownsampleThreshold 1.50000
/EncodeGrayImages true
/GrayImageFilter /DCTEncode
/AutoFilterGrayImages true
/GrayImageAutoFilterStrategy /JPEG
/GrayACSImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >>
/GrayImageDict << /QFactor 0.15 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >>
/JPEG2000GrayACSImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >>
/JPEG2000GrayImageDict << /TileWidth 256 /TileHeight 256 /Quality 30 >>
/AntiAliasMonoImages false
/DownsampleMonoImages true
/MonoImageDownsampleType /Bicubic
/MonoImageResolution 1200
/MonoImageDepth -1
/MonoImageDownsampleThreshold 1.50000
/EncodeMonoImages true
/MonoImageFilter /CCITTFaxEncode
/MonoImageDict << /K -1 >>
/AllowPSXObjects false
/PDFX1aCheck false
/PDFX3Check false
/PDFXCompliantPDFOnly false
/PDFXNoTrimBoxError true
/PDFXTrimBoxToMediaBoxOffset [
0.00000
0.00000
0.00000
0.00000
]
/PDFXSetBleedBoxToMediaBox false
/PDFXBleedBoxToTrimBoxOffset [
0.00000
0.00000
0.00000
0.00000
]
/PDFXOutputIntentProfile (U.S. Web Coated \050SWOP\051 v2)
/PDFXOutputCondition ()
/PDFXRegistryName (http://www.color.org)
/PDFXTrapped /Unknown
/SyntheticBoldness 1.000000
/Description << /FRA
/JPN
/DEU
/PTB
/DAN
/NLD
/ESP
/SUO
/ITA
/NOR
/SVE
/ENU
>>
>> setdistillerparams
<< /HWResolution [2400 2400] /PageSize [612.000 792.000] >> setpagedevice

The Mismeasure of Crime

Mosher, Clayton; Miethe, Terance D.; Hart, Timothy C.

CHAPTER 3

OFFICIAL CRIME DATA

“The statistics of crime and criminals are known as the most unreliable and difficult of all statistics. First, the laws which define crimes change. Second, the number of crimes actually committed cannot possibly be enumerated. This is true of many of the major crimes and even more true of the minor crimes. Third, any record of crimes, such as arrests, convictions, or commitments to prison, can be used as an index of crimes committed only on the assumption that this index maintains a constant ratio to the crimes committed. This assumption is a large one, for the recorded crimes are affected by police policies, court policies, and public opinion.

—Sutherland (1947, p. 29)

Official crime data are those that derive from the normal functions of the criminal justice system. These official counts of crime include police reports of offenses and arrests, charges filed by prosecutors, criminal complaints and indictments, imprisonment data, and prison releases.

Although official data come from a number of different sources, both the volume and nature of recorded crime incidents change dramatically through successive stages of criminal justice processing. A funnel analogy is often used to describe how both the number of offenders and the number of criminal offenses decreases significantly as one moves from police statistics to imprisonment data. Of all offenders and offenses known to the police, only a portion is subject to arrest. Only some of those subject to arrest will be prosecuted in courts, and of those, only some will be convicted. An even smaller proportion will be incarcerated. The most inclusive official measure of crime thus involves police reports of criminal incidents.

This chapter examines the nature and scope of police statistics on crime. We begin with a description of the crime reporting procedures in the United States. We then summarize historical trends in crime rates and the characteristics of offenders that derive from police reports and proceed to consider the various problems associated with using police data as a measure of crime. The chapter concludes with a discussion of cross-national data on crime.

UNIFORM CRIME REPORTS IN THE UNITED STATES

As discussed in Chapter 2, prior to 1930, police reports of crime in the United States were not collected or compiled in any systematic way across jurisdictions. Some large cities kept yearly counts of reported crime incidents and persons arrested, whereas other cities did not formally record such information. The classification of crime also varied widely across jurisdictions, with different community standards and legal definitions affecting how crimes were defined and whether particular activities were recorded as crimes in official data. Public tolerance and law enforcement activities toward lynching, abortion, spouse abuse, drug and alcohol use, dueling, and other forms of mutual combat varied widely both within and between southern and northern states. Both comparisons across jurisdictions and estimates of historical trends in crime are extremely hazardous prior to 1930 because of the lack of uniformity in definitions of crime and in the collection of police data on crime incidents.

In developing the Uniform Crime Reporting (UCR) program in the late 1920s, the International Association of Chiefs of Police (IACP) recognized that not all crimes are equally important. They therefore focused on seven types of crime that were prevalent, generally serious in nature, widely identified by victims and witnesses as criminal incidents, and most likely to be reported to the police. The original seven major index crimes, or what are also referred to as Part I offenses, include murder and manslaughter, forcible rape, robbery, aggravated assault, burglary (both commercial and residential), larceny, and motor vehicle theft. The reporting of other offenses (referred to as Part II or nonindex offenses) is not mandatory for police departments that participate in the UCR program. A list of Part I and Part II offenses is presented in Exhibit 3.1.

Although the number of police departments participating in the UCR program increased over time, the program remained essentially unchanged in its content and structure from its inception in 1930 until 1958. During that period, the FBI published crime data according to the size of the jurisdiction and did not provide reports of a national rate of crime because there was insufficient coverage of the entire country. Changes in 1958 included (1) the use of a composite crime index of all Part I offenses in the UCR, (2) the elimination of negligent manslaughter and larceny under $50 as Part I crimes, (3) the removal of statutory rape from UCR counts, and (4) the estimation and publication of crime rates for the entire United States.

Further changes to the UCR program, involving the development of state level officials to serve as intermediaries between local police departments and the FBI, were implemented in the 1970s. There are currently 47 states with special UCR programs that provide technical assistance within their state and submit data to the federal UCR program. The number of law enforcement agencies reporting to the UCR has almost doubled since the introduction of these state programs.

In 1979, arson was added to the UCR crime index as a Part I offense. This was in response to an apparently growing problem with this crime. In the United States in 1977, arson was reported to account for approximately one quarter of all fires and “perhaps about 750 deaths and possibly many more” (Simpson, 1978). Senator John Glenn (1978) was instrumental in having arson classified in the UCR, noting, “A criminal could steal a car in New York and drive it to New Jersey and his crime would be noted in the FBI charts. But let that same criminal torch a house or business—causing untold property damage and ruined lives—and his crime of arson will never make the charts. That’s a ridiculous situation” (p. 15). Despite protestations of FBI officials who believed it would be difficult to properly classify arson incidents in the UCR (Renshaw, 1990), Glenn’s argument that including arson as a Part I crime would focus national attention on a solution to the problem ultimately held sway.

The most fundamental change in the UCR program in the last three decades involves the movement toward what is known as a national incident-based reporting system (NIBRS), the special features of which will be addressed later in this chapter.

Although participation in the UCR program is voluntary, the proportion of law enforcement agencies that submit data to the program is remarkably high. A total of nearly 18,000 state, county, and city law enforcement agencies, covering more than 288 million inhabitants, submitted crime reports under the UCR system in 2008. A total of 95% of the U.S. population is covered by this data source, with participation rates slightly lower in cities outside metropolitan areas (88%) and in rural areas (90%).

DATA COLLECTION PROCEDURES UNDER THE UNIFORM CRIME REPORTS PROGRAM

Crime data under the UCR program are collected on a monthly basis from participating local law enforcement agencies, and they are typically submitted to a centralized crime records facility within their state UCR program. These completed crime report forms are then returned to the FBI for purposes of compiling, publishing, and distribution (FBI, 2004).

A national reporting system such as the UCR that relies on the cooperation of local and state agencies requires the development and establishment of standard operating procedures and uniform practices. Accordingly, the FBI has gone to considerable lengths to standardize these reporting procedures through the provision of training services and data collection manuals to local agencies.

According to the Uniform Crime Reporting Handbook (FBI, 2004), basic minimum standards in several areas are required for agencies providing data for the UCR program. First, a permanent written record is made of each crime immediately upon receipt of a complaint or a call for service. A follow-up system is used to examine whether reports are promptly submitted in all cases. Second, crime reports are checked to see that all offenses submitted in the UCR program conform to the UCR classification of offenses. Third, all records and statistical reports are closely supervised by the agency administrator. Periodic inspections are made to ensure strict compliance with the standard rules and procedures.

CLASSIFYING AND SCORING CRIMINAL OFFENSES IN THE UCR PROGRAM

Two essential components of the UCR data system involve the classifying and scoring of criminal offenses. Classifying crime offenses in the context of the UCR refers to the process of translating offense titles used in particular local and state laws into the standard UCR definitions for Part I and Part II offenses. Depending on the particular classifications used in individual jurisdictions, this conversion process may be more or less ambiguous for certain offenses. Scoring of criminal offenses, in contrast, refers to counting the number of offenses after they have been classified under the UCR typology and entering the total count on the appropriate form. Uniformity in both classifying and scoring criminal offenses across jurisdictions is essential for maintaining the integrity of the UCR.

The Uniform Crime Reporting Handbook (FBI, 2004) provides reporting agencies with detailed definitions and general rules for the classification and scoring of criminal offenses. The classification of offenses into particular UCR categories is based on the facts that underlie an agency’s investigation of the crime. The UCR program distinguishes between crimes against persons (i.e., criminal homicide, forcible rape, and aggravated assault) and crimes against property (i.e., robbery, burglary, larceny-theft, motor vehicle theft, and arson). Under the UCR scoring rules, one offense is counted for each victim in crimes against persons and one offense is counted for each distinct operation in crimes against property. Motor vehicle thefts are an exception to the property-counting rule in that one offense is counted for each stolen vehicle.

Given that UCR definitions of criminal offenses are a crucial element in the standardization of reporting practices, it is important to look more closely at how major criminal offenses are defined and counted under the UCR scheme. As described in the 2004 UCR Reporting Handbook and the methodological appendices for all publications of the Uniform Crime Reports, the Part I offenses are defined as follows:

Criminal homicide involves two subtypes of offenses. Murder and nonnegligent manslaughter are defined as “willful (nonnegligent) killing of one human being by another” (p. 15). The second type of criminal homicide involves manslaughter by negligence, which is defined as “the killing of another person through gross negligence” (p. 18).

Forcible rape is defined as “the carnal knowledge of a female forcibly and against her will” (p. 19). It involves two categories: (a) rape by force and (b) attempts to commit forcible rape. These offenses are restricted to female victims, and they are classified as forcible regardless of the age of the victim. Nonforceable offenses against victims under the age of consent, fondling, and incest are excluded.

Robbery is defined as “the taking or attempt to take anything of value from the care, custody, or control of a person or persons by force or threat of force or violence and/or by putting the victim in fear” (p. 21). Robbery involves a theft or larceny but is aggravated by the element of force or threat of force.

Aggravated assault is defined as an “unlawful attack by one person upon another for the purpose of inflicting severe or aggravated bodily injury” (p. 23). This type of assault is usually accompanied by the use of a weapon or by means likely to produce death or great bodily harm. Simple assaults are excluded.

Burglary is “the unlawful entry of a structure to commit a felony or theft” (p. 27). Attempted forcible entry is included in this category.

Larceny-Theft involves the “unlawful taking, carrying, leading, or riding away of property from the possession or constructive possession of another” (p. 31). Larceny-theft is subclassified into the following categories: (a) pocket picking (i.e., theft from a person by stealth), (b) purse snatching that involves no more force than necessary to snatch the purse from the person’s custody, (c) shoplifting, (d) thefts of articles from motor vehicles, (e) thefts of motor vehicle parts and accessories, (f) thefts of bicycles, (g) thefts from buildings, (h) thefts from coin-operated devices or machines, and (i) all other larceny-theft not specifically classified. Attempted larcenies are included in this category.

Motor vehicle theft is the theft or attempted theft of a self-propelled vehicle that runs on land surface and not on rails. Motorboats, construction equipment, airplanes, and farming equipment are specifically excluded from this category.

Arson involves “any willful or malicious burning or attempt to burn, with or without intent to defraud, a dwelling house, public building, motor vehicle or aircraft, personal property of another, etc” (p. 37). Fires of suspicious or unknown origin are excluded from the UCR.

Sources of Ambiguity

Coding crimes into these categories can be a complex process. As previously noted, the FBI provides training to local reporting agencies and presents numerous examples in the UCR Reporting Handbook (FBI, 2004) to illustrate the rules for classifying and scoring criminal offenses. However, there are several sources of ambiguity in the definition and coding of even the UCR Part I offenses that call into question the uniformity of reporting practices across jurisdictions. In fact, it is not unreasonable to assume that all Part I offenses are subject to considerable variability in counting and scoring across individual reporting units. The primary sources of variability include differences across local jurisdictions in their interpretation of crime incidents, the hierarchy rule, the diligence of record keeping, and the adequacy of follow-up procedures.

In the specific case of homicide, the main obstacle to uniform reporting and counting involves the follow-up procedures, the timing of police investigations and UCR filing, and definitional ambiguity in the classification of accidental killings and justifiable homicides. For example, the recording of situations of aggravated assaults that become murders because the victim dies as a result of the assault assumes equal diligence and detailed record keeping across reporting agencies in conducting follow-up investigations and correctly adjusting multiple monthly returns. Some less reliable agencies may simply count the aggravated assault and fail to record the subsequent death of the victim as a murder.

An interesting example of the confusion that can be created with respect to the coding of homicides comes from New York City, which, in 2006, saw an apparent increase in homicides from the previous year. However, part of this increase was fueled by an unusual number of deaths that were classified as homicides because the city’s medical examiner determined they were related to crimes that had been committed in earlier years. Of the 25 such reclassified deaths in 2006, 12 were related to injuries that had occurred at least 14 years earlier, including one case of a 72-year old man who has shot in 1974 and died of pneumonia in April of 2006 (Vasquez, 2006).

Depending on when in the investigative process the UCR incident is filed, a deadly shooting involving two juveniles playing with a gun may be classified as accidental (i.e., manslaughter by negligence) or willful killing (i.e., murder and nonnegligent manslaughter). Similarly, the killing of an individual by a law enforcement officer or private citizen in the course of the commission of a felony by that individual is a justifiable homicide under the UCR, but some local agencies violate UCR procedures and count such incidents as criminal homicides. Such differences in classification are not likely to be identified in the record-checking procedures used by the FBI.

The major source of ambiguity in the definition and classification of forcible rape involves what constitutes “carnal knowledge of a female forcibly and against her will.” Specifically, some jurisdictions may apply the strict definition of carnal knowledge as sexual intercourse (i.e., penilevaginal intercourse), whereas others may consider a fuller range of sexual acts and offensive touches. Also, when there is no apparent resistance on the part of the victim, some jurisdictions may count the act as consensual and, thereby, not against the woman’s will. Contrary to the instructions provided in the UCR forms, local reporting agencies may also vary in their inclusion of male victims and female offenders in their counts of forcible rape. There is also likely to be considerable variation across local jurisdictions in the inclusion and counting of forcible rapes that occur within the context of marital partners and intimates.

Sources of diversity in the classification of robberies are related to the distinction between strong-arm robberies and types of larceny from the person (e.g., purse snatchings). Under the UCR classifications, a purse snatching is classified as a strong-arm robbery when force or threat of force is used to overcome the active resistance of the victim. This force is also considered more than is necessary to snatch a purse from the grasp of the person. However, is it reasonable to assume that all local law enforcement agencies and, for that matter, individual police officers share the same interpretation of “more than necessary” force? Likewise, if the victim falls to the ground when a bag or purse is yanked from her shoulder, would this offense be classified uniformly as robbery or as larceny-theft? Does the classification change if the victim was pushed rather than falling or stumbling to the ground? In addition, jurisdictional differences are likely in the counting of robberies with multiple victims in the same behavioral incident. The UCR rule is to ignore the number of victims and count “one offense for each distinct operation” (FBI, 2004, p. 10), but can we be certain that this rule is uniformly applied? How is this rule actually applied across jurisdictions in cases of spree robberies that may be interpreted as a continuation of the original incident?

Definitional and classification problems with aggravated assault concern the interpretation of the provision that it is not necessary that physical injury results from an aggravated assault. Threats and assaults in the context of domestic violence are also subject to various interpretations. When assault situations occur in private places with no witnesses besides the victim, the absence of physical injuries makes it especially difficult to ascertain on a consistent basis whether an aggravated threat or attempt with a dangerous weapon actually occurred. The mere brandishing of a dangerous weapon may also be interpreted by some, but not by other local agencies, as an aggravated assault. Domestic assault situations are especially problematic in their classification across jurisdictions. Physical injuries to victims of domestic violence are often treated under state codes as gross or simple misdemeanors rather than felonies such as aggravated assault. Whether a threat with a dangerous weapon was involved (or a weapon was merely brandished) is also difficult to uncover in this particular context. Even under the best conditions of training and definitional clarity, local agencies will vacillate widely in their UCR classification of offenses with threats or no physical injury as aggravated or simple assault.

The major obstacles to uniformity in classifying and scoring the crime of burglary are the demonstration of intent beyond unlawful entry, the inclusion of attempts, the types of persons who qualify as being involved in an unlawful entry, and more general definitional misunderstandings. For example, burglary is a trespass with intent to commit a felony or theft, but how is this intent consistently determined when the alleged burglary is only attempted and not completed? Could the incomplete act be just a trespass, the destruction of property, or a type of vandalism? Does the apprehension of a suspect after breaking a window count as an attempted burglary or simply vandalism? Concerning the difference between lawful and unlawful entry, are acts of theft without forcible entry by previous intimates (e.g., ex-spouses, separated but not divorced parties, ex-roommates) counted as burglaries or larcenies? This determination will vary depending on the interpretation of particular parties as having the necessary legal status to define their behavior as lawful entry.

McCleary, Nienstedt, and Erven (1982) examined some additional problems in the classification of burglary. In an interview with a UCR coding clerk in a particular police department, they were informed that “a burglary has the element of breaking and entering a building. In a lot of cases, the thief breaks through a fence and steals something. That’s not a burglary, but a lot of officers don’t know that” (p. 362) and would still classify such an incident as a burglary.

Another issue with respect to the classification of burglary stems from what is known as the hotel rule. Under this rule, “if a number of units under a single manager are burglarized and the offenses are most likely to be reported to the police by the manager rather than the individual tenants/renters, the burglary should be reported as a single incident” (FBI, 2004, p. 62). Examples include burglaries of a number of hotel rooms or storage units in commercial self-storage buildings. Note that under this rule, even though a number of separate burglaries may have occurred, only one would be recorded in official data.

The major problem with the classification of larceny-theft stems more from the differential likelihood across jurisdictions of reporting particular types of thefts than from definitional ambiguity. Specifically, police underreporting and undercounting of particular thefts, such as shoplifting and stolen motor vehicle parts or accessories, is especially likely when these offenses involve minor financial losses and occur in large metropolitan areas. These frequently occurring offenses, however, may be more accurately reported to the FBI in smaller local areas.

Another, perhaps more obvious, problem with the larceny-theft category is related to the estimate of the dollar value of the item(s) stolen. The dividing line for UCR reporting was $50, and larceny more than $50 was the index offense that increased the most over the early history of the UCR—an increase of more than 550% between 1933 and 1967. However, because the purchasing power of the dollar in 1967 was only 40% of what it was in 1933, many thefts that would have been under $50 in 1933 were more than $50 in 1967 (President’s Commission on Law Enforcement, 1968).

Differences across local areas in the UCR counting and scoring of motor vehicle theft may derive from the lack of internal consistency in the coding of motor vehicle thefts and thefts of accessories and parts. Although the UCR manuals clearly specify the different categories, it is possible that some agencies may assume that the theft of motor vehicle accessories and parts falls into the category of motor vehicle theft rather than larceny-theft. The theft of boats and bicycles may also be improperly classified as motor vehicle theft by some local jurisdictions.

The differential interpretation of “willful” or “malicious” burnings and how suspicious fires are classified are the major problems associated with the UCR category of arson. The Uniform Crime Reporting Handbook clearly notes that suspicious fires of unknown causes should not be counted as arson. However, local areas are likely to vary widely in their investigative expertise in these crimes and their subsequent reporting of fires as arson.

The FBI has gone to considerable lengths in an attempt to monitor the accuracy of classifying and scoring crimes in the UCR. Starting in 1997, the FBI developed a voluntary Quality Assurance Review (QAR) for the UCR program that assesses the validity of crime statistics through an on-site review of local case reports. The review program also extends to the collection and compilation of crime statistics by the state repositories. Upon completion of the review, the QAR assessment team sends the agency a written evaluation of its performance in reporting methods, submission requirements, and overreporting or underreporting of incidents. Each state’s UCR program is subject to a QAR evaluation at least once every three years to determine the level of compliance with national UCR guidelines (FBI, 2008).

The Hierarchy Rule and Counting Multiple-Offense Incidents

The UCR’s hierarchy rule applies to the classification and scoring of crimes when multiple offenses are committed at the same time by a person or group of persons. When the hierarchy rule is applied in a multiple-offense situation, only the most serious offense in the series is reported, and all others are ignored. For example, if an individual breaks into a house, steals items from the house, kills the owner of the house, and makes a getaway in a stolen car, only the murder will be recorded in official statistics. Similarly, if, during the commission of a robbery, the offender strikes the teller with the butt of a handgun, runs from the bank, and steals an automobile at curbside, it would appear that three Part I offenses (robbery, aggravated assault, and motor vehicle theft) have occurred. However, because robbery is the most serious of the three offenses, only it would be counted; the two other offenses would be ignored (FBI, 2004).

The hierarchy rule, in theory, involves the application of a rather simple two-step process. First, the reporting agency classifies each of the separate offenses and determines which of them are Part I crimes. Second, the ranking of Part I crimes under the UCR system is used to identify the most serious offense, and that offense is recorded in the data. The decision to apply the hierarchy rule becomes more complicated when it is unclear whether there was a separation of time and place between the commissions of several crimes.

The major methodological concern regarding the hierarchy rule is how to determine compliance with it and what adjustments, if any, should be used to correct for potential classification errors. Greater oversight by the state or federal UCR program is an obvious way of determining compliance, but such coding decisions are usually not visible or detectable because the summary counts provided in monthly UCR data do not include the information necessary to make independent judgments of coder reliability. Perhaps ironically, the most direct solution to the problem of selective application of the hierarchy rule is its elimination through the greater utilization of the National Incident-Based Reporting System.1

NATIONAL INCIDENT-BASED REPORTING SYSTEM

A recent enhancement to the UCR program is the development of an incident-based reporting system for reporting offenses and arrests, known as the National Incident-Based Reporting System (NIBRS). It is described as “a new approach to measuring crime, one that is simultaneously ambitious, revolutionary, cumbersome, little known, and disappointingly slow to be adopted” (Maxfield, 1999, p. 120). Implementation of the NIBRS program requires (a) a revision of the definitions of certain offenses, (b) the identification of additional significant offenses to be reported, and (c) the development of incident details for all UCR offenses (see FBI, 1997). When fully implemented, it is believed that NIBRS data will be better able to measure the true volume of crime than standard UCR data because the former does not rely on the hierarchy rule and other practices that restrict the counting of crime incidents.

In contrast to the traditional UCR, which uses a summary or aggregate reporting approach, NIBRS categorizes each incident and arrest in one of 22 basic crime categories (see Exhibit 3.2 ) that span 46 separate offenses. A total of 53 data elements about the victim, property, and offender are collected under NIBRS (see Exhibit 3.3). Both Barnett-Ryan (2007) and Addington (2007) provided a fuller description of the NIBRS approach and its conceptual and methodological comparability with the UCR’s traditional summary reporting system and the NIBRS data.

NIBRS was intended to be implemented as a phase-in program, and it has largely developed at that pace. The FBI has accepted NIBRS data from local agencies since January 1989. During its first 10 years of the program, the FBI certified a total of 19 state-level programs for participation in NIBRS. As of February 2008, 31 states have been certified for NIBRS. Three additional states have individual agencies submitting NIBRS data, and other states remain in the testing or development stage. Five states (Alaska, Florida, Georgia, Nevada, and Wyoming) have no formalized plan to participate in NIBRS (see www.jrsa.org/ibrrc/background-status/nibrs_states.shtml).

Although NIBRS data have been used in federally published reports on crime, it remains too early to determine the overall effectiveness of this alternative UCR program. As of February 2008, only about 25% of the U.S. population is covered by NIBRS reporting. Participation in the program has also been concentrated within small- and medium-sized law enforcement agencies. In fact, none of the current agencies that report NIBRS data cover a population of 1 million residents or greater. The low rate of participation in NIBRS by agencies with populations of greater than 250,000 residents has raised serious questions about the accuracy of national estimates of violent crime that derive from NIBRS data (see Addington, 2007).

Despite its promise in terms of improving the accuracy of crime measurement, several potential problems exist with NIBRS data. Most obvious is the incredible complexity of the coding schemes: The coding specifications are documented in four volumes published by the FBI. As Maxfield (1999) suggested, few police officials are researchers, and diligence in paperwork is not among the skills most valued by police officers. As a result, missing data may become an even greater problem under the NIBRS because of the larger number of categories for which data are collected and the complexity of definitions within each of these categories. Furthermore, as Roberts (1997) noted, the incentives for law enforcement agencies to participate in NIBRS data collection are few. These agencies may feel that NIBRS data are of far more value to researchers than to themselves, and there is concern that the detailed, incident-level reporting required for NIBRS will require police officers to spend additional time filling out reports instead of responding to the needs of the public. A widespread perception also exists that NIBRS participation will result in an increase in reported crime because the UCR’s hierarchy rule will be eliminated. This presents a potential public-relations disaster for law enforcement agencies that are, to at least some extent, evaluated on the basis of crime rates in their jurisdiction.

OFFICIAL CRIME TRENDS AND PATTERNS BASED ON UNIFORM CRIME REPORTS

One of the primary purposes for the establishment of uniform crime reporting practices across jurisdictions was to provide a national barometer of crime and its distribution. The methods of classifying and counting offenses have remained relatively stable over time, allowing for estimation of national crime trends. Aggregate characteristics of particular types of offenses and some demographic characteristics of arrested persons are also presented in these national statistics.

Based on UCR data, the crime rate in the United States has vacillated over time and exhibits some variation by type of crime. Participation in the UCR program was sufficient to estimate national crime trends beginning in the 1960s. Starting then, the total crime rate per 100,000 inhabitants increased steadily until the mid-1970s, then decreased somewhat, and then peaked again in the early 1980s. It generally rose steadily from the mid-1980s to the early 1990s and has generally dropped since that time (see Exhibit 3.4).

Although the number of reported crimes exceeded 11.1 million in 2008, the crime rate of 3,667 per 100,000 is at the lowest point since 1968; the crime rate has declined by about 14% over the last 10 years (1999 to 2008). Declining crime rates are found in each region of the country. Southern and western states have continued to experience the highest rates of reported crime, and lower rates are found in the northeast and Midwest.

NOTE: Violent crimes include murder and nonnegligent manslaughter, forcible rape, robbery, and aggravated assault. Property offenses include burglary, larceny-theft, and motor vehicle theft. Arson is not including in the data presented.

The FBI’s Crime in the United States, 2008 (FBI, 2009) indicated that violent crime (i.e., murder, rape, robbery, and aggravated assault) accounted for about 12% of the total Part I offenses reported to law enforcement, whereas the remaining 88% were property crimes. This ratio of violent to property crimes in national data has been quite stable over time. Throughout the history of UCR reporting, larceny-theft represents, by far, the most common offense in these national data, whereas murders are the least common offense. Aggravated assaults account for about 60% of all violent crimes.

Homicide

Among all violent crimes, the most comprehensive police data are collected on murder and manslaughter. This is the case because (a) as the most serious UCR offense, this crime is never undercounted by the hierarchy rule; (b) murder has the highest clearance rate of all index crimes (i.e., 64% of murders known to the police in 2008 were cleared or “solved” by an arrest); and (c) additional police data are collected on each homicide through the Supplementary Homicide Reports (SHR).

Both the number of homicides and rate per 100,000 population have followed a similar pattern to the trend for all Part I offenses combined. Homicide rates increased throughout the 1960s until the mid-1970s, dropped somewhat in the late 1970s before the peak appeared in 1980, stayed relatively high until the early 1990s, and have decreased steadily since that time. In 2008, 16,272 homicides were known to the police, representing a 5% decline from 1999. The homicide rate of 5.4 per 100,000 population in 2008 was the lowest recorded in the United States since the mid-1960s.

Homicide rates based on UCR data vary across geographical areas. The homicide rate in southern states (6.6 per 100,000 population) is higher than in any other region of the country. However, each region has experienced a declining homicide rate over the last five years. Cities within large metropolitan areas had a 2008 murder rate of 6 per 100,000, compared to rates of about 3 per 100,000 for non-metropolitan areas. Homicide rates in particular U.S. cities over time, however, exhibit fairly unique patterns. Some cities have homicide rates that have fluctuated considerably between 1960 and 2000 (e.g., Houston), some cities have stable rates over this period (e.g., Baltimore, Phoenix, Seattle), and others have experienced general increases with dramatic upward swings in a particular decade (e.g., Detroit, New Orleans, Washington, D.C.).

Exhibit 3.5 reveals considerable variation in homicide rates across major U.S. cities in 2008, with a low of 4.8 per 100,000 in Seattle to a high of 63.6 in New Orleans.

Analysis of the FBI’s Supplementary Homicide Reports (SHR) for 2008 indicates several dominant patterns in the characteristics of homicide victims and offenders (see Exhibit 3.6). More than three fourths of homicide victims are males, and nearly 9 out of every 10 victims are aged 18 years or older. Almost half of all homicide victims are black. Black males have the greatest risk of being homicide victims of all sex-race combinations. Concerning offender characteristics, approximately 90% of homicides for which complete information was available are comprised of male offenders, and the vast majority of homicide offenders (91%) are persons aged 18 years or older. In 2008, about half of all homicide offenders were black, and the clear majority of homicides (87%) were intraracial killings. Males are most often murdered by male offenders (91%), and about 90% of female homicide victims are killed by males.

Based on police reports of known offenses, homicides also exhibit wide variation in their offense characteristics and situational contexts (see Exhibit 3.7). The majority of victims know their assailants, and most of these incidents involve killings by acquaintances or friends (47%) or a family member or intimate partner (31%). The killer is a stranger in approximately one in five murders (22%) in which information about the victim-offender relationship is known. Arguments and disputes are the most prevalent circumstances under which homicides take place (56%), and a sizable minority (31%) of killings occur in the course of the commission of another felony offense (especially robberies). A firearm was the most common lethal weapon used in homicide incidents; 67% of homicides in the SHR data involved the use of a firearm, whereas approximately 13% involved knives or other cutting instruments. The proportion of homicides involving the use of firearms has changed very little over the last 30 years.

Forcible Rape

Based on UCR data, rape rates in the United States increased steadily and more than tripled between the early 1960s and the early 1980s, remained high and fairly stable across the 1980s, and have generally decreased since that time. An estimated 89,000 rapes were known to the police in 2008, representing a rate of about 29.3 per 100,000 inhabitants.

Rape rates also vary by location. States in the northeast have considerably lower rape rates than other regions of the country. Forcible rape rates in metropolitan areas are far higher than those found in non-metropolitan areas.

Several other factors are associated with rape in the UCR data. Most rapes known to the police involve completed offenses by force (92%), whereas the remaining cases involve attempts. According to UCR arrest data, about 43% of those arrested for forcible rape in 2008 were under the age of 25, and about one third of arrestees for this crime were black.

Robbery

Robbery rates in the United States have vacillated widely over the last 40 years. These rates more than quadrupled from 1960 to 1980, then dropped in the early 1980s, rose dramatically in the late 1980s until 1991, decreased appreciably until the year 2000, and have remained relatively stable since that time. More than 440,000 robberies were known to the police in 2008, representing a rate of 145 per 100,000 inhabitants. Comparative UCR data indicate that the estimated number of robberies in the United States has decreased by approximately 36% from 1991 to 2008.

Similar to other violent crimes, robbery incidents vary by geographical location (see Exhibit 3.8). Southern states have the highest robbery rates of all regions, and the Midwest has the lowest rate. Although nearly half of all robberies in the United States are street muggings, a far higher proportion than the national average for muggings is found in the northeast. Convenience store robberies account for the highest proportion of robberies in the South. Although when the general public thinks about robberies they most likely envision bank robberies, these robberies account for only about 2% of all robberies, and they are rare in all geographical areas. Robbery rates are highest in the largest metropolitan areas, and street muggings account for a large proportion of robberies in such jurisdictions compared to other areas.

Several other characteristics of robbery are revealed in UCR data. For example, the average monetary loss from a robbery is approximately $1,300, which ranges from $712 taken in robberies of convenience stores to $4,854 for the average bank robbery. With respect to the type of weapon used in robberies, firearms (44%) are the most common, followed closely by strong-arm tactics (40%). Males accounted for about 9 out of every 10 robbery arrestees, and nearly two thirds of persons arrested for this crime were under 25 years of age. More than half of arrested robbers are black.

Compared to the UCR data from the early 1970s, there has been both change and stability in the factors associated with robbery over time. Robbery rates across this time frame have remained higher in major metropolitan areas than other locations, and similar proportions of robberies are found to involve strong-arm tactics over time. Based on arrest data, a similar proportion of robberies across time periods involve males, but the prevalence of robbery arrests for persons under 25 years old has decreased somewhat over time (76% in 1972 vs. 66% in 2008).

Aggravated Assault

Aggravated assault rates increased in almost every year from 1960 to the early 1990s, before decreasing rather steadily over the next 15 years. More than 830,000 aggravated assaults were estimated from UCR data in 2008. The estimated rate of 275 aggravated assaults per 100,000 population is the lowest recorded since 1978. Southern and western states have the highest rates for this offense, and rates of aggravated assault are more than twice as high in large metropolitan areas than in non-metropolitan counties.

Concerning offense and offender characteristics, the most common weapons used in aggravated assaults are blunt objects (34%) and personal weapons such as hands, fists, and feet (26%). Knives or cutting instruments (19%) and firearms (21%) are the other types of weapons used in these assaults. Males accounted for about 79% of those arrested for aggravated assaults in 2008, and approximately 40% of these arrestees were under the age of 25. Although the largest majority of aggravated assault arrestees were white (63%), black offenders represented a higher proportion of persons arrested for aggravated assault than their distribution in the U.S. population (34% of arrestees vs. 12% of the population).

Property Crime

Property crimes account for about 88% of the Part I offenses known to the police. As a group, property crime rates increased dramatically between the 1960s and early 1980s, vacillated up and down for the next 10 years, and then exhibited a steady decline after 1991. More than 9.7 million of these offenses were known to the police in 2008. The property crime rate of 3,212 per 100,000 in 2008 represented the lowest rate since 1972. Both property crime rates and incidents are highest in southern states and lowest in the northeast (see Exhibit 3.9). These rates and incidents are also far higher in large metropolitan areas than other locations.

Burglary rates in the United States more than doubled between 1960 and 1980 and have generally declined since the early 1980s. More than 2.2 million burglaries were known to the police in 2008. The burglary rate in southern states is more than double the rate in northeastern states (941 vs. 430 per 100,000). Metropolitan areas also have far higher rates and incidents of burglary than non-metropolitan areas. The vast majority of burglaries involve forcible entry (61%), and residential break-ins account for over two thirds of all burglaries (70%). The majority of burglaries occur during the daytime hours, whereas the majority of nonresidential burglaries happen at night. Males, persons under 25 years of age, and blacks are overrepresented among burglary arrestees.

Larceny-thefts are the most common crime in the UCR data, involving an estimated 6,957,412 offenses in 2008. Rates of larceny increased throughout the 1960s and 1970s, peaked and remained high in the 1980s and mid 1990s, and have dropped by about 30% since that time. Southern and northeastern states have the highest and lowest larceny rates, respectively. Larceny-theft rates are higher in cities within and outside metropolitan areas than in nonmetropolitan counties.

The average value of property loss due to larceny in 2008 was $925. The average take from pocket picking was $563, and losses from purse snatchings were approximately $427, compared to $196 for shoplifting and $1,540 for thefts from buildings. Thefts from motor vehicles are the most common type of larceny-theft, accounting for about 26% of these crimes. Only about 1% of larceny-thefts involve either pocket picking or purse snatching. More than half of the arrestees for this offense are under 25 years old, and about 29% are black. Females were arrested for this offense more often than for any other Part I offense, comprising 41% of larceny-theft arrestees.

Motor vehicle theft rates based on UCR data increased throughout the 1960s, hovered between 400 and 500 per 100,000 in the early 1970s to the mid-1980s, and similar to the trend with other crimes, have decreased since the 1990s. Nearly 1 million auto thefts were known to the police in 2008, and the theft rate was substantially higher in western states compared to other regions. Large urban areas have offense rates that are far higher than smaller cities and non-metropolitan counties. The average value of the stolen vehicle was approximately $6,751. Seventy-two percent of stolen vehicles were automobiles, 18% were trucks or buses, and the remainder were other types of vehicles. Arrestees for motor vehicle theft are disproportionately male (83%), under 18 years old (25%), and black (38%).

Although only about 80% of the U.S. population is covered in UCR estimates for arson for 2008, several patterns are revealed in these data. First, the overall arson rate is estimated to be 24 per 100,000 population in 2008. These estimated rates are far higher in cities with a population greater than 250,000 than in smaller locations. Second, most arsons known to the police involve physical structures (43%), followed by mobile property (29%) and other types of property (28%). Third, the average dollar loss per arson offense was about $16,000. Industrial and manufacturing structures had the highest average losses of $212,000 per offense. Fourth, arrested arsonists are disproportionately male (84%), juveniles under 18 years old (47%), and black (22%).

Hate Crimes

Hate crimes were added to the UCR in 1990 with the passage of the Hate Crime Statistics Act. Hate crimes, also known as bias crimes, are defined as offenses committed against a person, property, or society that are motivated, in whole or in part, by the offender’s bias against a race, religion, sexual orientation, ethnicity/national origin, or disability (FBI, 2008). The agencies that participated in the FBI’s Hate Crime Statistics Program in 2008 covered about 89% of the U.S. population within 49 states and the District of Columbia.

Based on this UCR program for Hate Crime Statistics, a total of 7,783 hate crime incidents involving 9,168 offenses were reported by agencies in 2008. Nearly all of these incidents involved a single type of bias. The most common motivation reported in these incidents was bias based on race (51%), religion (20%), sexual orientation (17%), ethnicity or national origin (12%), and disability (1%). Most of the offenses in racial-bias hate crime incidents were motivated by anti-black bias (73%), and the next largest group within these incidents involved anti-white bias (17%). The majority of hate crimes based on religion were motivated by anti-Jewish bias (66%),whereas anti-male homosexual bias (59%) was the primary motivation for hate crimes directed at sexual orientation. Across all types of hate-crime bias, the majority of offenses involved crimes against persons (60%) rather than their property (39%). The criminal acts of intimidation (49%), simple assault (32%), and aggravated assault (19%) were the most common types of crimes against persons in these hate-motivated incidents.

Although hate crime statistics have been used and improved over the last two decades, there remains considerable doubt about the reliability of hate crime statistics for a variety of reasons. For example, the FBI emphasizes that the presence of bias alone is insufficient for determining that a crime is actually a hate crime, and they caution that a criminal incident should be reported as a hate crime only upon sufficient evidence from law enforcement agencies. Unfortunately, this principle does not guarantee that uniform coding is used across participating jurisdictions. For example, in the FBI’s 2006 report of hate crimes, one city (Washington, D.C., with 64) reported more hate crimes than at least 10 entire states, and the southern states of Alabama and Mississippi, both with long histories of racial tension, reported one and zero crimes, respectively (Fears, 2007; see also http://bjs.ojp.usdog.gov/content/pub/pdf/hcrvp ). In addition, about 84% of the agencies that participate in the UCR hate crime reporting program reported no such crimes in 2008. This concentration of hate crimes within a relatively small number of agencies is attributable to two possible explanations: (1) Hate crimes are highly concentrated within particular jurisdictions across the country or (2) classifications of hate crimes are unreliable and selectively used across jurisdictions. Given that both of these explanations are reasonable, estimates of the prevalence and nature of hate crime in the United States based on these data must be viewed with considerable caution.

Clearance Rates

One measure of the effectiveness of local law enforcement agencies in apprehending criminals is the clearance rate. Crimes are cleared by either an arrest of a suspect or by exceptional means when some element beyond the control of law enforcement (e.g., the death of a suspect, international flight) precludes them from making formal charges against the offender (FBI, 2008). As is often illustrated by the arrest of serial killers, the arrest of one person may clear several crimes. Alternatively, several people may be arrested in the clearance of one crime. Clearance rates represent the proportion of crimes known to the police that lead to arrest.

Clearance rates as reported in UCR data have varied over time, region of the country, and across the various types of Part I offenses (see Exhibit 3.10). According to the most recent UCR data, only 21% of the Part I offenses were cleared in 2008. Clearance rates were higher for violent crimes (45%) than for property crimes (17%). Among the types of violent crimes, murders had the highest clearance rate (64%), and the lowest rate was for robbery (27%). Larceny theft had the highest clearance rate among property crimes (20%), and the lowest rate was for motor vehicle theft (12%). The northeast had the highest clearance rates for all regions of the country for both violent and property offenses.

When examined over time, clearance rates for some crimes have decreased more rapidly than others. There has been more than a 25-percentage-point decline in clearance rates between 1960 and 2008 for murder and forcible rape in the United States. Wellford and Cronin (2000) noted that a number of factors affect clearance rates for homicide. For example, the probability of clearance increases significantly when the first police officer on the scene quickly notifies the homicide unit, medical examiners, and the crime laboratory and attempts to immediately locate witnesses and secure the area. The greater difficulty in identifying offenders because of the rise in the number of violent crimes involving strangers over the last two decades is another explanation for the declining clearance rates for these crimes. Smaller decreases over time are found in clearance rates for aggravated assault, burglary, motor vehicle theft,and robbery. The clearance rate for larceny-theft has remained at 20% across this time period (see Exhibit 3.10).

From the perspective of solving crime and distributing justice to offenders, the decline in clearance rates over time is troubling because it suggests that crime control efforts that focus on the punishment of offenders are increasingly ineffective due to the fact that a growing number of offenders are not subject to arrest. This situation is even more dire when one considers that there is considerable pressure on police departments to inflate their clearance rates for various political reasons. This inflation and distortion of clearance rates has long been recognized by critics of official estimates of crime (Kitsuse & Cicourel, 1963), but the actual nature and magnitude of these distortions in current police practices is largely unknown. Thus, a reasonable conclusion is that clearance rates provide only a gross representation at best of the potential solvability of crimes known to the police. Given differences in police recording practices and the nature of crime across different geographical areas, attempts to assess the effectiveness of various police departments by comparing their clearance rates for particular offenses is a questionable practice that has little or no scientific utility.

PROBLEMS WITH POLICE DATA ON CRIME

Police reports are often considered to be the best official measure of the nature and extent of crime. Compared to prosecutorial, judicial, and correctional data, police reports are more comprehensive in their coverage of types of criminal offenses and include information on criminal incidents even when the offender has not been identified. However, as a measure of the true extent of crime in a jurisdiction, police statistics are inadequate for several fundamental reasons. The major problems with police data involve variation in citizen reporting and police recording practices, possible race and social class biases in the structure of policing, limited coverage of crime types under UCR data, conceptual and methodological factors that affect the classification of crime incidents and estimates of national crime rates, and political manipulation and fabrication of these data by police departments and other reporting agencies.

Variation in Citizen Reporting and Police Recording Practices

As discussed earlier, the term dark figures has been widely used by criminologists to represent the gap between the true extent of crime and the amount of crime known to the police. The primary sources of this gap are the inability of police to observe all criminal activity, the reluctance of crime victims and witnesses to report crime to the police, and variation in the recording of known crime incidents due to police discretion.

Contrary to the image portrayed in crime dramas and media depictions of police work, the vast majority of crime becomes known to the police through citizen complaints or calls for service. In other words, police mobilization toward crime and its detection is largely because of a citizen complaint. If a member of the public fails to contact the police about a criminal incident they have experienced or witnessed, it will remain undetected in most cases. The magnitude of unreported crime vastly exceeds crime reported to the police.

The reasons victims and other citizens do not report most crimes to the police are wide and varied (see Hart & Rennison, 2003). Some victims lack trust in the police or have severe reservations about the ability of law enforcement officials to solve crimes. Some fear retaliation and reprisals from offenders for reporting crimes; others think it is not worth their while to report offenses because, for example, the property is uninsured and probably will not be recovered. The victims in some crime situations may also be involved in criminal activities themselves (e.g., drug sellers or prostitutes who are victims of robbery), which decreases their likelihood of reporting. Others may believe the incident was a private matter, that nothing could be done, or that it was not important enough. Public apathy and the desire to not get involved may underlie some witnesses’ reluctance to report offenses they observe. Regardless of the particular reasons for underreporting of crime by citizens, this reporting gap raises serious questions about the accuracy of police data as a valid measure of the prevalence of crime.

Even if a crime incident is reported by citizens or directly observed by the police, there is no guarantee that such an offense will be recorded in police data. In fact, police discretion both across and within jurisdictions in recording an incident as a crime is a major source of inconsistency in official counts of crime. In this context, the role of the police dispatcher can be crucial. Pepinsky (1976) found that the decision of patrol officers about whether to report offenses was determined by the nature of the calls they received from the dispatcher. Apparently, if the dispatcher named no offense in the call or dispatched the officer to check a victimless or attempted offense, the chances were practically zero that the officer would report an offense.

In his classic study of police-citizen encounters, Donald Black (1970) identified the following factors that determine whether an incident reported by citizens is formally recorded as a crime by the police:

Legal Seriousness of the Crime

Police are more likely to write up a crime report when the crime is more serious. Approximately 72% of the felonies but only 53% of the misdemeanors in Black’s study were written up as reports. This means that the police officially disregarded about one fourth of the felonies they handled.

The Complainant’s Preferences

When called to a crime scene, police often follow the wishes of the complainant. They almost always agree with the complainant’s preference for informal action (as opposed to arrest) in minor cases. When the complainant requested official police action, the police complied in the majority of both felony (84%) and misdemeanor (64%) situations.

The Relational Distance

Police are more likely to file an official report in cases involving strangers rather than friends or family members. Black (1970) asserted that the victim-offender relationship is more important than the legal seriousness of the crime in terms of whether an incident is officially recorded.

The Complainant’s Deference

The more deference or respect shown to the police by the complainant, the more likely it is that the police will file an official crime report. This pattern was found for both felony and misdemeanor situations.

The Complainant’s Status

Police are more likely to file an official report when the complainant is of higher social status. The effect of the race of the complainant on recording practices in Black’s study was unclear.

Differences in citizen reporting and police recording practices are also likely to vary by region of the country and by rural and urban jurisdictions. It is for these reasons that statistics on crime incidents are highly suspect for comparisons across jurisdictions.

Race and Social Class Biases in Policing

There is considerable evidence of racial and social class biases in street-level policing, which dates back to the earliest studies of police in the United States (see, e.g., Chicago Commission on Race Relations, 1922; Myrdal, 1944; Sellin, 1928). Irwin (1985) argued that a tendency on the part of police to characterize lower-class persons and blacks as disreputable and dangerous may lead them to watch and arrest such individuals more frequently than is warranted on the basis of their actual criminal involvement. Although focused more explicitly on socioeconomic status as opposed to race, Sampson (1986) provided further evidence of police bias in arrest decisions. In a study examining the police processing of juveniles in the Seattle, Washington area, Sampson found that for the bulk of offenses committed by juveniles, official police records and referrals to court were structured not simply by the act itself but by the socioeconomic and situational contexts of such acts. In addition, law enforcement officials apparently perceived lower-class neighborhoods as being characterized by a disproportionate amount of criminal behavior and accordingly concentrated their patrol resources in those “offensible space” areas (Hagan, 1994). As Smith (1986) suggested,

Based on a set of internalized expectations derived from past experience, police divide the population and physical territory they must patrol into readily understandable categories. The result is a process of ecological contamination in which all persons encountered in bad neighborhoods are viewed as possessing the moral liability of the area itself. (p. 316)

Under these conditions, it is possible that at least some of the difference between minority and white crime rates is the product of a differential police focus on minority groups (Mosher, 2001).

Limited Coverage of Different Crime Types

Police statistics on crime such as those developed under the UCR system are restricted to only a small class of criminal offenses. Most of these crimes involve street-level offenses that occur among individuals. UCR data do not measure federal crimes or political crimes, and they severely undercount organizational and occupational crime. Corporate crimes such as price-fixing and environmental pollution are simply not covered by these data, and occupational crimes such as thefts and frauds by employees are underrepresented in UCR data. Beirne and Messerschmidt (2000) contended that there are at least three reasons why the FBI focuses on crimes committed by the powerless: (1) the FBI recognizes the fact that crimes typically or exclusively committed by the powerful are difficult to detect, often covered up, and seldom reported to the police; (2) the FBI is insensitive to the plight of the powerless; and (3) the FBI is politically biased in favor of the powerful.

Conceptual and Methodological Problems

Police data on crime in the United States are also problematic as valid and reliable measures of crime prevalence because of several conceptual and methodological problems. As described in detail earlier in this chapter, the major conceptual problems involve the definition of certain crimes under the UCR and the classification of a particular offense under one of the included crime categories. Even with extensive coding and classification rules, counting and scoring decisions in practice are subject to multiple interpretations and potentially large inconsistencies both within and across jurisdictions. Basic methodological problems involve estimating population figures in order to calculate crime rates in noncensus years, sampling error, imputation and estimation procedures, and the application of the hierarchy rule and other conventions in cases of multiple crime incidents.

Estimating Population Figures to Calculate Crime Rates

The UCR calculates crime rates per 100,000 population; however, the most accurate counts of population are only available for census years (i.e., 1980, 1990, 2000, 2010, etc.). In noncensus years, estimates of the population are used to calculate crime rates; and if these estimates are inaccurate, then calculated crime rates will be similarly inaccurate. For example, Bell (1967) noted that 1949 crime rates for California, which were based on 1940 population figures, were grossly inflated because the state’s population increased by more than 3 million people over the decade. When the crime rate “automatically dropped … [in 1950] it was not due to sunspots or some other cyclical theory, but to a simple statistical pitfall” (p. 153).

Another example of estimation problems involves crime statistics for Illinois in 1999. In particular, the Illinois State Police report for 1999 underestimated the population of Chicago by tens of thousands of residents, which produced an inflated crime rate for the city. An Illinois sheriff whose county’s crime rate was overstated claimed, “Hell, they never could add. You get those fellows off of chipped roads and they get confused” (as quoted in Berens & Lighty, 2001).

Sampling Error and Participation Rates

Participation in the UCR is voluntary, and police departments are not under any legal obligation to report their crime data to the FBI. The reporting area covered by the UCR program has remained high since the late 1950s. For example, the national coverage rate was 93% in 1972, 95% in 1999, and 95% in 2008. Active participation in the UCR program is highest among law enforcement agencies in large metropolitan areas and is lowest in cities outside metropolitan areas.

Sampling error is a problem in any research when sample data are used to estimate and represent population values. Two general sources of sampling error and possible sampling bias are found in the UCR system. First, not all police agencies in the United States report crime data to the UCR program. If there are differences in the crime experiences of reporting and nonreporting agencies (as is suggested by the differences in crime rates and participation rates by urban and rural areas), this sampling error is actually sampling bias that may distort population estimates. Second, agencies that are defined as participating may not be providing complete crime data to the FBI. In fact, data from six states were excluded in the 1997 UCR because of erratic or nonreporting behavior. A study of reporting behavior covering the years 1992 to 1994 revealed that only 64% of law enforcement agencies reported crime for the entire 36 months; 17% were classified as partial reporting (i.e., 1 to 35 months of data) and 19% provided no reports (Maltz, 1999). Given these conditions of incomplete reporting on the part of law enforcement agencies, claims that UCR data represent more than 90% of the U.S. population are misleading.

Incomplete reporting under the UCR program is due to a wide variety of reasons. Some of these include (a) natural disasters that prevent state agencies from submitting their data on time, or at all; (b) budgetary restrictions on police and the cutback on services; (c) changes in the personnel who prepare local UCR data and their replacement with persons with less training, experience, or commitment to the program; (d) new reporting systems or computerization of old systems that may cause delays or gaps in the crime reporting process; (e) small agencies with little crime that may feel it is unnecessary to file monthly reports; and (f) incompatibility in state and UCR definitions, resulting in data being submitted by states but not accepted by the FBI (Maltz, 1999). Whatever the reason, incomplete reporting and nonreporting have obvious implications for the estimation of national trends in crime.

Problems with Imputation and Estimation

Problems related to sampling error and potential sampling bias are compounded when estimating arrest trends and the profile of persons arrested for crimes. As noted earlier, clearance rates vary widely according to the type of crime, hovering around 50% for violent crimes and around only 17% for property crimes in 2008. Given that the majority of offenders are not counted in arrest data, inferences about the typical profile of particular types of offenders from UCR arrest data also represent a type of sampling bias because some offenders (e.g., nonstrangers) are more easily identified by victims, and subsequently arrested, than others. Another problem with developing offender profiles from UCR arrest data is that arrests are a reflection of differential police priorities and enforcement practices, further contributing to the likelihood of qualitative differences between those arrested and not arrested for even the same type of offense.

Since 1958, the FBI has used two different methods of imputing crime data for police agencies that have incomplete data or that do not provide reports at all (see Barnett-Ryan, 2007). If a particular agency reports for three or more months in a given year, the total annual crime for the jurisdiction is estimated by multiplying the reported number of crimes by 12, divided by the number of months reporting (Maltz, 1999). This procedure implicitly assumes that the crime rate for nonreporting months is the same as for reporting months, which is a rather dubious assumption—especially given that research has demonstrated that property crimes generally peak in the fall and winter months and violent crimes peak in the summer months (Baumer & Wright, 1996; Hird & Ruparel, 2007). If, on the other hand, an agency reports for less than three months, the number of crimes in that jurisdiction is essentially estimated from scratch. Such agencies are considered to be nonreporting agencies, and the FBI estimates data for these jurisdictions based on crime rates for the same year for similar agencies. These similar agencies are defined as those in the same population size category in the same state but that provide 12 months of data. If there are no comparable agencies in the state, the estimate is based on rates of crime in the jurisdiction’s region.

Unfortunately, if the nonreporting agency is different from the “comparable” reporting agencies on crime-related correlates other than geographical location and size (e.g., income distribution, unemployment rates, population density, racial composition), the assignment of equal proportions of crime in each jurisdiction will distort the accuracy of these estimates. The fact that no two cities are alike in their economic opportunity, physical structure, history, and culture raises questions about this estimation approach. And although it is possible that inaccuracies in crime data that result from such estimation procedures may not be significant, the real problem is that there is currently no way of determining whether the estimation procedures produce major or minor discrepancies in crime data. As Maltz (1999) pointed out, such imputation can be especially problematic for crimes that vary according to season.

Alternative imputation methods have also been used with UCR data. For example, the process of conversion to the NIBRS program required the estimation of totals for some entire states. Unique estimation procedures are also required when yearly data for a particular jurisdiction are incomplete and in other situations (e.g., the inability of some state UCR programs to provide forcible rape figures in accordance with UCR guidelines). For these problems, the UCR program has used known data from other geographical areas in the same time period, regional data from the United States for that year (e.g., mountain states, west north-central division), or state totals from previous years to derive population estimates. Such extrapolations, however, are accurate only if trends in other jurisdictions or the same jurisdiction in previous years are representative of crime experiences in the nonreporting areas.

Although it is often overlooked by UCR-data users, the UCR program has relied extensively on extrapolations from other jurisdictions or other time frames for estimating national crime trends. The most recent UCR report (FBI, 2008) provides the following examples of major nonreporting and estimation practices over the last 10 years.

Several states over various years did not report valid UCR Part I offense counts. Crime trends from previous years or from other states in their geographic division were used to calculate estimates of current trends. These estimation procedures have been used in Kansas (1998–2000), Kentucky (1998–2003), Illinois (1998–2008), Maine (1999), Montana (1998–2000), New Hampshire (1998–1999), and Wisconsin (1998).

Some State UCR programs did not provide forcible rape figures in accordance with UCR guidelines. These states included Illinois (1998–2008) and Minnesota (2005–2008). Forcible rape totals were estimated for these states using national rates within eight population groups (e.g., cities with more than 250,000 population, cities with 100,000 to 249,999 population, suburban counties, rural counties) and then assigning counts of forcible rape proportional to each state’s distribution in these population groups.

 

From these examples of imputation of UCR data, it is clear that cross-jurisdictional and over-time comparisons must be made with considerable caution.

Political Manipulation and Fabrication

An additional limitation of official crime statistics involves their manipulation and fabrication for political purposes. For better or worse, police departments are evaluated to some extent on the basis of the volume of crime in their jurisdiction. The mass media, city and county commissions, local chambers of commerce that promote tourism in their “safe” city, elections for incumbent police chiefs and sheriffs, and the general public are sources of considerable pressure on police departments to provide a positive spin on the effectiveness of their crime fighting activities. Although Chambliss (1984) suggested that “other things being equal, it is in the interests of the police to prove an increase in crime [because] higher crime rates … mean increased budgets” (p. 176), the image of a rising crime rate is not generally good news for local businesses and police departments that are held accountable for these crime trends. Favorable crime statistics apparently make everyone happy. In the early 1970s, for example, several large police departments in the United States downgraded their crime rates “to create the illusion that the country is a safer place to walk at night because President Nixon’s anti-crime measures are working” (Justice Magazine, 1972, p. 1).

In another example of this manipulation, Seidman and Couzens (1974) identified a significant decrease in the number of larceny-thefts of $50 or more in one jurisdiction as a result of the installation of a new police chief who threatened to replace police commanders who were unable to reduce the amount of crime in their precincts. The importance of the $50 criterion was that larcenies of less than $50 were not reported to the FBI. Thus, simply by estimating the value of stolen goods to be slightly less than $50, it was possible for the police to reduce the official crime rate. Similarly, McCleary et al. (1982) noted that a significant decline in the number of burglaries in one jurisdiction was related to a change in police procedure whereby detectives, as opposed to uniformed police officers, investigated burglary complaints. When this experiment of using detectives was terminated 21 months later, the burglary rate in the jurisdiction increased again. In another example, a 72% increase in the number of major crimes in New York City from 1965 to 1966 was primarily due to a change in crime reporting; the actual increase was only 6.5% (Weinraub, 1967). In a perhaps even more disturbing example, in 1973, Orange City, California, based the pay of its police officers on decreases in crime. At least partially as a result of this change, the reported crime rates for rape, robbery, auto theft, and burglary dropped by 19% in this jurisdiction over a one-year period (Holsendorph, 1974).

Given that the police have exclusive control over the dissemination of crime data and that there is little monitoring of the accuracy of their crime counts, one obvious way to demonstrate effective law enforcement is to distort, manipulate, and fabricate the number and nature of crime reports. The claim of “cooking the data” has long been alleged against law enforcement agencies, and numerous incidents of police misconduct over the last few decades have increasingly challenged the integrity of law enforcement and have led to growing suspicion about the pervasiveness of cooking data across the country.

When submitting crime data to the UCR program, there are various ways for local agencies to distort and undercount crime incidents. The most basic methods for creative accounting through falsifying crime reports include the following:

Not reporting all crime incidents on monthly UCR submissions

Combining separate events as if they occurred in multiple-offense incidents and falsely using the hierarchy rule to undercount the total number of crime reports

Declaring large numbers of reported crimes as unfounded so they are not counted in UCR annual summaries

Downgrading major Part I offenses to minor offenses so they are not tallied nationally in the UCR summaries

The particular reasons or motives for the police manipulation of crime statistics are wide and varied. Economic interests and political posturing are sometimes the underlying cause of the artificial inflation of crime statistics by law enforcement agencies, whereas these and other reasons may be the basis for the undercounting of crime. The following examples illustrate both the diversity of motives and the magnitude of distortion and manipulation of crime statistics by law enforcement officials.

Crime Reporting in Philadelphia, Pennsylvania

Some of the most serious allegations of fabrication of crime statistics involve practices in the Philadelphia Police Department. The distortion and manipulation of crime statistics in this jurisdiction has grown out of a history of statistical manipulation that goes back for decades. In 1953, for example, Philadelphia reported 28,560 index crimes plus negligent manslaughter and larceny under $50, which represented an increase of more than 70% compared to 1951 figures. This tremendous increase in crime, however, was not due to an “invasion by criminals” (Bell, 1967, p. 152) but to the discovery by the new administration that earlier crime records had minimized the amount of crime in Philadelphia for a number of years. In fact, one district in the city had actually handled 5,000 more complaints than it had recorded (President’s Commission on Law Enforcement and the Administration of Justice, 1968). This distortion of crime statistics apparently continued; in 1970, Philadelphia, which was the fourth largest city in the United States, reported fewer index crimes than any other city among the 10 largest. In fact, Baltimore, which had less than one half the population of Philadelphia, reported more than 60% more index crimes in 1970 (Seidman & Couzens, 1974).

The two major forms of distortion that have been employed by the Philadelphia police in more recent years are the excessive use of “unfounded” and downgrading. It is estimated that literally thousands of sexual assault cases that occurred in the 1980s and 1990s in Philadelphia were buried by the sex-crime unit either by rejecting many of them as unfounded or by placing nearly one third of its caseload into noncriminal categories, such as “investigation of person” and other “throwaway categories” (Faziollah, Matza, & McCoy, 1998). When the high rates of unfounded sexual assaults were scrutinized, the sex-crime unit reported low rates for the next year simply by shifting these cases to “investigation of persons,” which are excluded in police summary data reported to the FBI.

A number of different types of downgrading have been used in Philadelphia to circumvent the counting of Part I offenses. The city has consistently had one of the lowest rates of aggravated assault of any large city because many of these attacks are classified as “hospital cases” or are downgraded to simple assaults, thereby being excluded from UCR data on serious crimes. Similarly, burglary is often downgraded to “lost property,” car thefts and break-ins are redefined as “vandalism,” and street muggings without injury (categorized as robbery in the UCR) are downgraded to the minor offenses of “threats.”

The manipulation of crime statistics in Philadelphia has been so notorious that dramatic actions have been taken to explore its source and curtail the practice. These procedures have included the auditing of police crime figures by the city controller’s office, the assignment of 45 detectives to reinvestigate more than 2,000 sex offenses that were downgraded over a five-year period, the appointment of an academic panel to develop yearly auditing procedures, and a formal inquiry by former U.S. Attorney General Janet Reno. The Philadelphia police commissioner in the late 1990s took several measures to increase the accuracy of police data, including the dismissal of district captains who were in charge of crime data and the use of undercover investigators posing as crime victims to determine whether police are recording the incidents accurately.

Although these corrective actions should improve the accuracy of police statistics in this jurisdiction, the impact of these presumed improvements in reporting practices on actual crime rates in Philadelphia is debatable. In 1998, the Philadelphia police department failed to report an estimated 37,000 UCR Part I crimes, but when these crimes were included, Philadelphia moved from the fifth to the second most dangerous city in the United States (“Numbers,” 2000). Not surprisingly, police officials in Philadelphia attributed this major increase in crime rates to more accurate reporting rather than a surge in violent behavior. However, by blaming rising crime trends on better reporting procedures, officials in this jurisdiction may be engaging in other forms of manipulation and creative writing to deflect attention away from ineffective law enforcement practices. It is within this context that both accurate and inaccurate reporting of crime may be functional for local police departments.

Crime Reporting in Atlanta

In addition to the situation with the Philadelphia police force, there is also evidence that police officials in the city of Atlanta manipulated crime statistics through the use of the unfounded option. Slightly more than one rape report per week was written off by the Atlanta police as never having happened in 1996 (Martz, 1998). These rape reports and nearly 500 robberies were quickly classified as unfounded and not counted in official crime reports.

By eliminating these serious crimes from official records, the Atlanta police department was able to make their city appear less violent than it actually was that year. City officials claimed that rapes had declined by 11% from 1995 (when including the unfounded rapes would have revealed a 2% increase) and that robbery rates had declined by 9% (instead of increasing by 1% when the unfounded robberies were included). The timing of this downgrading of violent crime was crucial because the Olympic Games were held in Atlanta in late August of 1996 and a mayoral election was also held that year (Martz, 1999).

As a result of a major public dispute between the Atlanta police chief and her deputy chief who managed the crime statistics section of the department, a joint state-federal FBI audit was conducted to assess the accuracy of crime reports in the city. The audit revealed that approximately 16% of the cases examined in the mid-1990s were improperly classified as unfounded, providing support for the deputy chief’s allegation that the department was manipulating crime reporting (Martz, 1999). Based on UCR policies, the reporting error rate for the most serious violent crimes was 26% in 1996.

As might be expected, the Atlanta police chief blamed the high error rate on confusion in the UCR classification rules for unfounded cases, rather than on the department’s deliberate manipulation of the crime data. However, a former robbery detective was quoted as saying that detectives were encouraged in subtle ways by supervisors to record particular types of cases as unfounded (e.g., homeless people, both suspects and victims who were drug users). This detective noted, “The system was set up to cheat a little bit, not to cheat in big numbers” (as quoted in Martz, 1999).

The Crime Drop in New York City

Substantial reductions in official crime rates in New York City from 1995 to 2000 have been attributed to aggressive and effective law enforcement practices by police officials. However, critics have alleged that the reduction is due to distortion and manipulation through downgrading cases as unfounded.

The claims of manipulation of crime data in New York City have been supported by surveys of retired New York Police Department captains and higher-ranking officials. More than 100 of these retired officials said that intense pressure to produce annual crime reductions led some supervisors and precinct commanders to make “ethnically inappropriate” changes to complaints involving the UCR’s Part I offenses that helped portray their precincts and the entire city as a safer place (Rashbaum, 2010). The NYPD’s CompStat program, which was implemented in 1995 and is used to provide close monitoring of crime trends within precincts, is often identified as a major impetus for political manipulation of crime data because it helped establish the idea that precinct commanders would be held accountable for the level of crime in their areas. By holding them accountable, the CompStat system provided both the means and the incentive for dubious and questionable crime reporting and recording practices.

The specific ways in which official crime data were allegedly manipulated in this city include downgrading of offense severity and altering the nature of the victim’s criminal complaint. For example, by reducing the value of items stolen (e.g., by selectively using the lowest price of an item on eBay or in a catalog), felony grand larceny (over $1,000) that should be recorded as a Part I offense could be downgraded to misdemeanor theft, which are not recorded as a Part I UCR crime. Retired senior officers also cited examples of precinct commanders or aides they dispatched going to crime scenes to persuade victims not to file criminal complaints or to modify their accounts in ways that could circumvent its classification as a Part I offense (Rashbaum, 2010). The following comment by the recording secretary of the New York Patrolmen’s Benevolent Association provides further insight into the ways in which crime numbers were manipulated in this context:

“You eventually hit a wall where you can’t push it down anymore. So commanders have to get creative to keep the numbers going down. So how do you fake a crime decrease? It’s pretty simple. Don’t file reports, misclassify crimes from felonies to misdemeanors, undervalue property lost to crime so it’s not a felony, and report a series of crime as a single event. A particularly insidious way to fudge the numbers is to make it difficult or impossible for people to report crimes—in other words, make the victims feel like criminals so they walk away just to spare themselves further pain and suffering.” (as quoted in Moses, 2005)

Although periodic reports of cooking crime statistics suggest that the validity of these police reports remain questionable, the actual magnitude of data manipulation or fabrication under NYPD’s reporting practices is largely unknown. However, over the last several years, NYPD Commissioner Raymond Kelly has implemented an auditing system to maintain the integrity of the crime reporting system in his jurisdiction. This greater scrutiny of crime reporting involves auditing every precinct’s books twice a year, correcting and revising crime statistics that derive from any errors discovered in this process, and holding personnel accountable with disciplinary actions when these errors are due to intentional manipulation. While these actions are laudable, the low visibility of many street-level police practices fall outside of the purview of these auditing reviews.

Crime Reporting in Other Jurisdictions

There are numerous other jurisdictions that have been identified in media outlets for their questionable practices in the reporting of crimes. These include the following:

• Reductions in the violent crime rates reported by the Los Angeles Police Department over the last decade have been called into question because of the undercounting of aggravated assaults. This has occurred primarily by the classification of physical batteries involving injury or threat of serious injury as domestic violence offenses when they occur in this context (Orlov, 2006). Given the prevalence of these types of aggravated assaults, this jurisdiction appears to have dramatically decreased its rate of aggravated assault and violent crime rate in general by recording these crimes outside of the UCR category of Part I offenses.

• Reviewing police and medical examiner records, the Detroit News (LeDuff & Esparza, 2009) contended that the Detroit Police Department was systematically undercounting homicides, leading to a falsely low murder rate for the city. Their review indicated that the police department incorrectly reclassified 22 of its 368 slayings in 2008 as “justifiable” so they did not report them as homicides under the UCR standards for murder and manslaughter. The investigative reporters also found at least 59 of these omissions over the previous five years.

• Downgrading in Boca Raton, Florida, resulted in an 11% decline in the felony crime rate in 1997. In that jurisdiction, a police captain downgraded crimes reported by investigating officers as burglaries and car thefts to vandalism and suspicious incidents. In one particular case, the captain changed a burglary charge to vandalism when $5,000 worth of jewels was taken and $25,000 in damage was done (Rozsa, 1998).

• The use of categories such as vandalism, trespassing, or missing property to downgrade residential burglaries has also occurred in smaller cities such as South Bend, Indiana (Sulok, 1998). Alternative strategies include delaying the submission of crime data until after elections, as has occurred in some cities. Political opponents allege that these delays are used to conceal potentially damaging crime trends, whereas the incumbents claim the delays are due to such factors as computer problems that prevent the timely release of the data.

• In 2008, a 22% increase in the most serious violent crimes was reported for England and Wales. However, Home Office officials attributed this increase to inaccurate record keeping—at least 13 of the 43 police forces in England and Wales had previously been classifying assaults involving grievous bodily harm with intent as less serious violent assaults (Travis, 2008).

• In 2006, Japan’s reported rate of intentional homicides was 0.44 per 100,000 (United Nations Office on Drugs and Crime, 2009), one of the lowest per capita homicide rates in the world. However, in that country, autopsies are performed in only 11% of all cases of unnatural death, and it has been alleged that law enforcement authorities discourage autopsies that might uncover a higher rate of homicide in their jurisdictions and exert pressure on doctors to attribute unnatural deaths to health reasons. As Wallace (2007) commented, “Odds are that people are getting away with murder in Japan.”

Official data can also be distorted through the peculiar practices of individual police departments with respect to some crimes. For example, between 1996 and 2000, Detroit had arrested far more people in homicide cases than any other big-city police department, reporting an average of nearly three arrests per homicide. Most cities average roughly one arrest per homicide. As a result of these practices, Detroit, with less than 2% of the population of the United States, accounted for 1 in 13 homicide arrests in the United States in 1998 and 1999. When questioned about these statistics, officials in Detroit claimed that they were the result of computer glitches or the arrests of people at homicide scenes on unrelated charges (Belluck, 2001).

The pattern of manipulation, distortion, and fabrication of official crime data is a serious problem that may be self-perpetuating. For example, the considerable media attention devoted to the declining crime rate in New York City has placed great pressure on other cities to report similar reductions in crime. If these data are generated through selective reporting practices, however, this may persuade other jurisdictions to use creative counting methods. Given these factors, decreases in the number of UCR Part I offenses over time may be more reflective of changes in police reporting and recording activities than changes in criminal activity in the larger society.

The Serial Killer Epidemic

During the early and mid-1980s, considerable media attention was focused on the apparent serial killer epidemic in the United States. Riveting television interviews with serial killers such as Ted Bundy and Henry Lee Lucas helped arouse public hysteria about these types of offenders. U.S. Justice Department officials, extrapolating from data in the UCR’s SHR, claimed that as many as 4,000 persons were murdered by serial killers in the United States each year.

Taking issue with these claims, Jenkins (1994) argued that Justice Department officials grossly inflated their estimates of the annual number of serial murders. This major counting error was the result of the rather questionable assumption that all or most of the SHR murders classified as “motiveless/offender unknown” were the work of serial killers. Jenkins concluded from his extensive analysis of serial killers that such offenders are responsible for no more than 350 to 400 murders in the United States each year.

The manipulation of official crime data to create the image of a serial killer epidemic served several organizational goals for the Justice Department. Specifically, this apparent epidemic provided an immediate justification for a new Violent Criminal Apprehension Program at a new center for the study of violent crime at the FBI Academy in Quantico, Virginia. The dramatic rise in the popularity of crime profiling was also initially based on this alleged serial killer epidemic.

Official Data on Juvenile Gang Crime

Official estimates of the number of youthful gang members and gang crimes have skyrocketed in the United States over the last three decades.

Agencies such as the National Youth Gang Center estimate from surveys of police departments that in the late 1990s, there were nearly 31,000 gangs and about 850,000 gang members in the United States (see Bilchik, 1999). However, because there is no uniform procedure for removing files of inactive gang members, law enforcement agencies’ estimates of the number and age range of gang members in their jurisdictions are very likely to be artificially inflated. In addition, political pressures to deny or minimize local gang problems, as well as the countervailing tendency to exaggerate them in order to secure monetary incentives to fight gangs, play a role in distorting the statistics on gang membership (Snyder & Sickmund, 1999).

In a study of whether the law enforcement response to gangs in Nevada was commensurate with the magnitude of the gang problem, McCorkle and Miethe (2001) found that police statistics on gangs are seriously distorted. For example, contrary to the image of dangerous youthful offenders portrayed in the media and other sources, these researchers discovered that a large percentage of gang members included on police gang rosters were adults, and a large percentage were individuals who had not been charged with any criminal offense. Instead, they were persons who had been “field identified” as gang members because of their associates, style of dress, race, and geographical location.

The official image of gangs as violent hordes with guns and selling drugs is also inconsistent with the substantiated prosecutorial data collected in Nevada. Specifically, McCorkle and Miethe (2001) found that less than 10% of violent crimes and drug offenses filed in Las Vegas courts involved gang members as suspects. Although official police statistics in Las Vegas indicated a dramatic rise in gangs and gang members in the 1980s, this presumed rise in gang activity over time was not validated by a rise in gang prosecutions.

The results of this study of youth gangs are interpreted as representing a “moral panic.” From this perspective, police used selective crime statistics and counting procedures to convey the notion of gang crime as a clear and present danger to the community. This presumed threat from youth gangs was more imaginary than real, but the police used their official statistics on gangs as part of a justification for additional financial resources to increase the size of the gang unit and to pass a special bond issue that provided for more police officers.

OFFICIAL CRIME DATA IN INTERNATIONAL CONTEXT

Given increasing globalization and concerns with examining crime trends crossnationally, it is instructive to examine the collection of official crime data in other countries. As such, in this section we discuss issues surrounding the collection of crime data by the International Police Agency (Interpol) and the United Nations through their world crime surveys.

Interpol Data

Interpol was established in 1923, and it has collected international crime statistics since 1950. Created in 1999, the European Law Enforcement Organization (Europol) also serves to facilitate the sharing of crime information among the countries of the European Union. Interpol’s first report was issued in 1954 and included data for only 36 countries; subsequently, reports were issued every two years and every year since 1993, with the inclusion of crime data on a greater number of countries.

As an international data source, Interpol reports are published in four languages (Arabic, English, French, and Spanish) and include data on murder, sex offenses, serious assault, theft, aggravated theft (robbery and violent theft and breaking and entering), theft of automobiles, fraud, counterfeit currency offenses, drug offenses, and the total number of offenses recorded in crime statistics of the member nations.

Although data are provided for multiple countries, Interpol reports provide cautions about the appropriateness of using these data for comparing crime rates across countries. The primary reason these comparisons are of dubious value is because the statistics do not account for definitional differences in crime categories across various countries, the diversity of methods used by different countries in compiling the statistics, and changes in laws or data collection techniques over time. In addition, as revealed from the International Victimization Survey (also see Chapter 5), the proportion of crime not reported to law enforcement varies substantially across nations and across different types of crime. These comparisons across countries are also adversely affected by the fact that countries with greater access to telephones in their households tend to report higher rates of crime, and countries in which household insurance is more available similarly report higher rates of crime. As a result, crossnational comparisons using Interpol data are most appropriate for crimes of extreme violence such as homicide, which are most likely to come to the attention of the police (Mosher, 2005a).

United Nations Crime Surveys

An additional source of international data on crime is the United Nations Crime Surveys (UNCSs). Data from these surveys are accessible through the United Nations Office on Drugs and Crime website (http://www.unodc.org). Early in its history, the UN expressed interest in the collection of criminal justice statistics at the international level, with resolutions concerning this issue at the UN Economic and Social Council meetings in 1948 and 1951. However, with the exception of one limited cross-national crime survey conducted over the 1937–1946 period, international crime data were not collected systematically until the early 1970s.

The more recent UNCSs were initiated in 1977, covering five-year intervals from 1970 (as of 2010, 11 UNCSs had been conducted.) The original rationale for collecting international statistics was to provide social scientists with additional data to examine the correlates and causes of crime, and the first UNCS received responses from 64 nations, providing crime and other data for the period between 1970 and 1975. The second UNCS covered the years 1975–1980 and represented an explicit change in purpose, with the emphasis moving from a focus on the causes of crime to issues surrounding the operations of criminal justice systems cross-nationally. The 11th UNCS covers the years 2007–2008.

The UNCS includes information from each responding country on the combined police and prosecution expenditure by year; the number of police personnel by gender; the total number of homicides (intentional and unintentional); the number of assaults, rapes, robberies, thefts, burglaries, frauds, embezzlements, and drug-related crimes; the number of people formally charged with crimes; the number of individuals prosecuted and the types of prosecuted crimes; the gender and age of individuals’ prosecuted; the number of convictions and acquittals; the number sentenced to capital punishment and other sanctions; the number of prisoners, the length of sentences they received, and prison demographics. However, it is important to note that these data are not complete for all countries responding to the surveys for all years.

Problems in Using International Crime Data

When making comparisons of official crime data across countries using Interpol and UNCS data, it is important to be aware of a number of problems with these comparisons. The first problem involves differences in the categorization of criminal offenses across nations. For example, the Netherlands has no category for robbery—an uninformed examination of data from that country might conclude that no serious property crime occurs. Similarly, in contrast to most other countries, Japan classifies assaults that eventually result in the death of the victim as an assault instead of a homicide.

Criminologists generally agree that homicide is the most similarly defined crime across nations, whereas rape and sexual offenses are likely the least similarly defined. However, even when comparing the nature and prevalence of homicides across countries, caution must be exercised. For example, although the UNCS collects information on the total, intentional, and unintentional homicides for each participating country, there are extreme differences in the proportion of homicides classified as intentional across countries, with a range from 10% to 100%. This clearly indicates the use of diverse criteria across countries in defining intentional and unintentional homicides. In the Netherlands, many relatively nonserious offenses are first recorded by the police as attempted murders (e.g., a situation in which the driver of a car almost hits a pedestrian). At least partially as a result of such coding, the Netherlands reported to Interpol in 1990 a homicide rate of 14.8 per 100,000 population, which, on the surface, makes it appear to have had a higher homicide rate than the United States in that year. Upon closer examination, however, over 90% of the offenses coded as homicide in that year were attempts.

Another problem with international comparisons involves the counting of crimes during times of internal conflict. In fact, the classification of casualties as homicides can be particularly problematic in nations that are experiencing war, rebellion, or serious civil and political conflicts. For example, Rwanda in the 1994 Interpol report showed a total of exactly 1 million homicides, which translates into a rate of 12,500 homicides per 100,000 population. Clearly, this figure for homicides in Rwanda, which is overly exact and incredibly high, includes deaths resulting from war and civil conflict in that country (Mosher, 2005a).

Errors in international crime data and its interpretation can also result from problems with coding and the calculation of rates. For example, Belgium’s recorded homicide rate in the 1994 Interpol report was 31.5 per 100,000 population, a figure that was 25 times higher than the figure for previous years. However, it turns out that a zero was left off the reported population of Belgium when calculating the homicide rate, resulting in a rate (which was based on a total of 315 homicides, 195 of which were attempts) calculated based on a population of 1 million rather than the actual population of 10 million.

Although these weaknesses with international crime data should be kept in mind, the danger in emphasizing the problems involved in comparing crime data across countries is that an inference could be made that nothing can be gained from gathering and analyzing such information. On the contrary, although these data cannot be reliably used to rank countries in terms of their levels of crime, they are appropriate for assessing the direction of change in crime over time and across nations.

As noted above, cross-national comparisons of crime are probably most appropriate for offenses such as homicide. Exhibit 3.11 provides data on homicide rates for 2008 for a selected group of countries, grouped by region, and indicates tremendous variation in these rates. In general, Latin American and Caribbean countries exhibit the highest homicide rates, with four of the countries in that region (El Salvador, Honduras, Jamaica, and Venezuela) having 2008 rates in excess of 50 per 100,000 population. European, North American, and Oceanic countries have comparatively lower homicide rates, with the Russian Federation, at 14.2 homicides per 100,000 population, having the highest rate among these nations. Although Iceland’s population was approximately 320,000 in 2008, it is notable that there were no homicides in that country in 2008. It is also notable that the United States 2008 homicide rate of 5.2 per 100,000 is considerably higher than that of Canada and most other western industrialized nations.

SUMMARY AND CONCLUSIONS

This chapter examined police statistics on the nature and prevalence of crime. We discussed the definitions of criminal conduct underlying the UCR classification system, the problems associated with classifying and scoring crimes under this system, the nature and prevalence of crime based on official measures, and the major limitations of police data as an accurate measure of crime.

Throughout the process of reporting and recording official instances of crime, criminal definitions are socially constructed. In other words, each official count of crime requires some amount of interpretation and negotiation. Under the widely regarded UCR system in the United States, a crime report becomes part of the official data only after surviving the following five decision points: (1) someone must perceive an event or behavior as a crime, (2) the crime must come to the attention of the police, (3) the police must agree that a crime has occurred, (4) the police must code the crime on the proper UCR form and submit it to the FBI either directly or through their state data collection agency (who must also correctly submit it to the FBI), and (5) the FBI must include the crime in the UCR (Beirne & Messerschmidt, 2000, p. 39).

Of the many problems associated with police statistics on crime, charges of political manipulation and fabrication of these statistics are particularly insidious because they challenge the basic integrity of the data. Although more extensive auditing and monitoring may improve data quality, the processing of police crime data remains largely unavailable for public scrutiny and thus continues to be susceptible to creative accounting methods that may serve political ends. Under conditions of growing distrust of statistical data and numerous allegations about the downgrading of offenses, UCR claims regarding declining national crime rates and the characteristics of offenders derived from these data may best be viewed as tentative estimates that are rooted on rather shaky grounds.

10.1177/0886260503251130ARTICLEJOURNAL OF INTERPERSONAL VIOLENCE / July 2003Miller / AN ARRESTING EXPERIMENT

An Arresting Experiment
Domestic Violence Victim Experiences

and Perceptions

JOANN MILLER
Purdue University

This study looks at the experiences and perceptions that domestic violence victims
reported with Mills’s power model. The victims’partners were the primary research
participants in an arrest experiment. The following were empirically examined: the
occurrence of violence following suspect arrest, victim perceptions of personal and
legal power, victim satisfaction with the police, and victim perceptions of safety fol-
lowing legal intervention. Race and two victim resource measures (i.e., employment
status and income advantage) explained variance in perceptions of independence. A
police empowerment scale was used to measure legal power. It was found that arrest
affected the probability of reoccurring domestic violence. Suspect arrest and the vic-
tim’s perceptions of legal power were related to perceptions of safety following
police intervention. The study concludes with some implications for domestic vio-
lence research, programs, and perspectives.

Keywords: domestic violence; intimate partner violence; perceptions; victim; arrest

We analyzed the victim interviews that were conducted as part of a random-
ized domestic violence arrest experiment, in Dade County, Florida, that was
designed to examine how police responses affected the likelihood of reoccur-
ring violence. Domestic violence suspects, the primary participants in the
field experiment, were assigned to an arrest or to a no arrest condition. The
nature of their offenses, domestic violence, generated a second type of
research participant: The victims, like the suspects, were subjected to the
arrest experiment. We examined the arrest study from the victim’s perspec-
tive by analyzing the interviews that were conducted soon after police inter-
vention and 6 months later. We examined the victims’ reports of reoccurring
violence, their perceptions of power, and their subjective experiences follow-

695

Author’s Note: This study was sponsored by a Social and Behavioral Science Center Fellow-
ship, Purdue University, and a Fellowship in Law and Sociology, Harvard Law School. The
author is most grateful to R. Gartner, Jonathan Miller, G. D. Hill, and two anonymous reviewers
who made useful comments on earlier drafts.

JOURNAL OF INTERPERSONAL VIOLENCE, Vol. 18 No. 7, July 2003 695-716
DOI: 10.1177/0886260503251130
© 2003 Sage Publications

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

ing police intervention, including their feelings of safety and their satisfac-
tion with how the police responded to their preferences. We discuss this
study’s implications for domestic violence explanations and programs,
focusing on the importance of understanding the role of victim perceptions
and empowerment.

THE SPOUSE ASSAULT REPLICATION PROGRAM

For two decades researchers have used randomized or experimental
designs to study how police practices can decrease the probability, frequency,
and severity of reoccurring family or domestic violence (Davis & Taylor,
1997; Ford, 1991; Maxwell, Garner, & Fagan, 2001; Sherman, 1992). There
are few field experiments more controversial than the collection of six stud-
ies, sponsored by the National Institute of Justice, that are known as the
Spouse Assault Replication Program (SARP). Endorsed by feminist advo-
cates and crime control proponents alike, the earliest results were reported on
television and in major metropolitan newspapers. Most urban police depart-
ments in the United States, in response to the widely publicized initial experi-
ment, developed mandatory or preferred arrest policies for domestic vio-
lence, although some analysts issued sharp warnings of likely victim harms
and injuries (Sherman, 1992).

The initial experiment was fielded in Minneapolis, and five quasi-
replication studies were fielded in Omaha, Colorado Springs, Milwaukee,
Charlotte, and Metro Dade County, Florida. An Atlanta experiment was also
conducted, but the data were not made available to social science researchers.
All the SARP studies were originally designed to explain the specific deter-
rent effect of suspect arrest on reoccurring or repeated family or domestic
violence. (Some of the post hoc explanations of the empirical findings are
based on social control theories.) Various methods were used across the
SARP sites to assign domestic violence suspects to an arrest treatment group
or to a no arrest control group.

Maxwell et al. (2001) reported that the SARP studies generated at least
300 potential outcome measures. Most of them, collected at two or three
points in time, focused on the suspect’s reoccurring violence that was perpe-
trated against the same intimate partner. One key type of outcome measure
examined the number and types of violent events that occurred following
police intervention. Another important type examined time to failure, or the
amount of time between the initial police response and a police record of a
subsequent offense.

696 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

The domestic violence victims were interviewed in all the SARP experi-
ments, primarily to corroborate police reports or other records of suspect
behavior. The typical victim interview schedule was designed to measure
characteristics of the victim’s relationship with the suspect and get detailed
reports of violent events and threats. In two of the experiments, Omaha and
Dade County, interviewers asked the victims to disclose detailed reports of
their perceptions and feelings following police intervention. This study is
based on the victims of the Dade County experiment.

SARP Results

Results from all but one of the SARP experiments were reported, some-
times to the press and often in social science journal articles (Lempert, 1989).
Sherman and Berk (1984), architects of the original Minneapolis experiment,
found that “the arrest intervention certainly did not make things worse and
may well have made things better” (p. 269). Reports from the other experi-
ments were more cautious. Analysis of the suspect data from the Colorado
Springs experiment showed that arrest had no deterrent effect. Analysis of the
victim interview data, however, uncovered modest deterrence, especially
among employed suspects (Berk, Campbell, Klap, & Western, 1992). Pate
and Hamilton (1992) reported an interaction effect between arrest and sus-
pect employment status in the Dade County experiment, leading them to sug-
gest that “the deterrent effect of arrest is influenced by the informal sanctions
implicit in employment status” (p. 695). Perhaps worst of all, early reports
based on the Charlotte, Milwaukee, and Omaha experiments concluded that
arrest had either no deterrent effect, or an escalation of violence effect, by 6
months following police intervention (Dunford, Huizinga, & Elliott, 1991;
Hirschel, Hutchison, & Dean, 1992; Sherman et al., 1991).

Berk et al. (1992), Sherman (1992), Garner, Fagan, and Maxwell (1995),
Gelles (1993), Mills (1998), and Maxwell, Garner, and Fagan (1999) con-
ducted meta-analyses of some or all of the SARP studies and reached sharply
divided conclusions. Did arrest deter domestic violence? Gelles concluded
that “a more complete and sobering look at . . . [the arrest experiments] indi-
cates that the initial claim of the deterrent value of mandatory arrest policies
may well be the social science equivalent of cold fusion” (p. 578). His posi-
tion was challenged by Berk (1993a, 1993b), and more recently by Maxwell
et al. (1999), who reported a slight or modest relationship between arrest and
repeat offending.

Maxwell et al. (2001) pooled select data elements across all the SARP
sites to resolve the basic questions advanced by the six randomized arrest

Miller / AN ARRESTING EXPERIMENT 697

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

experiments. They found no evidence to conclude that arrest escalated
domestic violence. Arrest, they reported, had a small and, in some experi-
ments, a statistically nonsignificant effect on suspect behavior. Most sus-
pects, regardless of the type of police intervention, did not reoffend. All told,
researchers who have studied empirical findings across SARP sites have
reported that arrest, along with individual and social psychological attributes
and characteristics, differentially affected recidivistic, misdemeanor domes-
tic violence. Suspects who experienced shame as a consequence of arrest, at
work or in their communities, were less likely to reoffend. However, those
with relatively low “stakes in conformity” (Toby, 1957) were not likely to be
deterred by arrest. The SARP studies showed that “the size of the reduction in
repeat offending associated with arrest is modest compared with the effect of
other factors (such as the batterer’s age and prior criminal record) on the like-
lihood of repeat offending” (Maxwell et al., 2001, p. 2; see also Garner et al.,
1995).

Victim Reports

Mills (1998) and Stephens and Sinden (2000) challenged any attempt to
reach definitive conclusions from the SARP experiments: “It is the victims
who have the most to gain (or lose) from the current [arrest] trend . . . but we
know little about victims’ experiences . . . and their interactions. . . . Their
voices are needed” (Stephens & Sinden, 2000, p. 535). Thus, our research
was designed to complement the published SARP studies by focusing exclu-
sively on the victims of one of the arrest experiments. Specifically, it
advances the Pate and Hamilton (1992) study, and it takes a step in the direc-
tion called for by Stephens and Sinden. We studied the Dade County victims’
reports to examine how their objective and subjective experiences were
related to police intervention, including suspect arrest. Based on victim data,
we analyzed reports of reoccurring violence immediately following the ini-
tial police call and 6 months later. In addition, we examined perceptions and
subjective experiences that were related to the arrest experiment.

PERCEPTIONS OF POWER

Mills (1998) analyzed the publications resulting from the SARP experi-
ments and concluded that uniform and mandatory programs, such as the
mandatory arrest of all domestic violence suspects or no-drop prosecution,
fail to stop the violence and protect the victims. Women, controlled and

698 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

abused initially by their partners, can be victimized once again by a “one size
fits all” legal response that does not consider the unique person’s needs to sur-
vive episodes of domestic violence. Mills also argued that the victim’s power
can be enhanced by effective legal intervention that incorporates the individ-
ual’s requirements and preferences. The victim, empowered by appropriate
police and prosecutorial responses, can prevent revictimization.

We adapted Mills’s (1998, 1999) model to distinguish two types of power,
personal power and legal power, that domestic violence victims in the Metro
Dade arrest experiment perceived and could use to prevent or stop violence.
We conceptualized personal power as a person’s perceived control over eco-
nomic and social resources. We conceptualized legal power as perceived
empowerment in response to police intervention.

Personal Power

Mills (1998) defined personal power as the social actor’s sense of control
when dealing with others, including a domestic partner. We analyzed per-
ceived independence as an indicator or a proxy measure for personal power.
That is, we believe that the women who perceived high levels of independ-
ence, relative to others, perceived higher levels of personal power.

We hypothesized that personal power is a function of work and earned
income. We expected to find that employed women and those with an income
advantage within their intimate relationships had stronger perceptions of
independence compared to unemployed women or compared to those with an
income disadvantage. Furthermore, we hypothesized that levels of independ-
ence were related to domestic violence experiences. Women with higher lev-
els of personal power, we hypothesized, would be less likely to experience
repeated acts of domestic violence following police intervention.

Our research hypothesis pertaining to personal power and reoccurring
domestic violence was derived from the empirical studies that examine how
levels of personal resources, or the control over resources, can empower vic-
tims to prevent repeated violence (see, e.g., Gelles, 1993; Jasinski, 2001c;
Johnson, 1992; McCloskey, 1996; Miller & Knudsen, 1999; Teichman &
Teichman 1989). Being employed outside the home is a social resource,
whereas income advantage is an economic resource. Employed women, in
principle, have access to information, and to social resources such as friend-
ships or work networks, at a higher level than unemployed women. An
income advantage can give a woman greater control or access to the financial
or economic resources of a household.

Miller / AN ARRESTING EXPERIMENT 699

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

Legal Power

Mills (1998) defined court system or legal power as the victims’ “percep-
tions of their role in the court process” (p. 310). We believe that legal power is
similar to Ford’s (1991) “power by alliance.” Ford’s concept is based on his
findings from a domestic violence prosecution study (Ford, 1991; Ford &
Regoli, 1993, 1998). A domestic violence victim can form a partnership or an
alliance with a legal actor, a police officer or a prosecuting attorney, who con-
veys respect and a concern for her safety. The alliance itself can be a powerful
resource that victims can use to prevent violence. The threat to call an ally
who has the power of the state to dispense in response to a criminal code vio-
lation has a greater deterrent effect than the threat to call a stranger or a friend.
An ally in criminal justice can also provide information and connections to a
network of social service providers. Legal power, used by victims of domes-
tic violence, can prevent reoccurring violence. It can also mediate the effects
of arrest or other forms of police intervention, similar to how informal mech-
anisms of social control mediate the deterrent effects of arrest on the suspects
(Pate & Hamilton, 1992; Sherman, 1992; Toby, 1957).

Legal power represents the woman’s perceived ability to control criminal
justice decisions and their consequences. We hypothesized that legal power,
regardless of whether the suspect was arrested, increased when the police
took legal actions that corresponded to the victim’s preferences. Further-
more, we hypothesized that victims’ subjective responses to police interven-
tion were related to their perceptions of legal power. We expected to find that
women who were satisfied with what the police did perceived higher levels of
legal power. Finally, we expected to find that women who perceived higher
levels of legal power following a police response experienced greater percep-
tions of personal safety.

RESEARCH METHODS

A Randomized Field Experiment

The data we analyzed are from the victim interviews that were conducted
as a part of the Dade County, Florida, arrest experiment from the SARP. The
principal investigators of the experiment designed the study to explain how
legal and informal sanctions deter misdemeanor domestic violence perpetra-
tors from repeated acts of abuse or violence (Pate & Hamilton, 1992; Pate,
Hamilton, & Annan, 1994). Whereas Pate and Hamilton (1992) focused on
the suspects and the consequences of formal and informal social controls,

700 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

this study focused on the victims. Thus, our research was designed to com-
plement Pate and Hamilton’s work. We examined empirically what the police
did in response to a domestic violence call, characteristics of the victim and
the suspect, the victim’s perceptions of personal and legal power, and victim
reports of domestic violence following police intervention.

The Dade County arrest experiment, conducted over a 3-year period, used
a unique, two-assignment design. Police randomly assigned each case to an
arrest or to a no arrest condition. Independently, they randomly assigned 50%
of the cases to a police Safe Streets Unit for counseling and follow-up investi-
gation. We examined arrest and Safe Streets assignment as two types of
experimental conditions that could influence recidivistic domestic violence.
Arguably, close police follow up, the hallmark of the Safe Streets Program, is
like intensive probation that is used to prevent recidivistic criminal behavior.

Of the assigned responses (arrest versus no arrest), 90% were actually
delivered in the Dade County experiment. The misassignment rate, or depar-
tures from the treatment or control group assigned, was higher in Dade
County than it was in some of the other sites (e.g., Milwaukee or Omaha) but
lower than it was in Charlotte (13%). Across all the SARP sites, the average
misassignment rate was approximately 3% (Maxwell et al., 2001). Sherman
(1993) argued that the misassignment rate, albeit considerably higher than
the ideal, does not severely challenge the internal validity of the study.

A total of 50.4% of the suspects were arrested, as assigned, and a total of
39.5% of the suspects were not arrested, as assigned. A correlation analysis
(not reported here in table form) showed that departures from the conditions
assigned were not related to the following characteristics, which have been
found in previous research to be related to the occurrence of domestic vio-
lence: a woman’s pregnancy, employment status, marital status, or race or
ethnicity (see, e.g., Jasinski, 2001a, 2001b; McCloskey, 1996; Straus &
Gelles, 1986). Likewise, personal and relationship characteristics were not
correlated with the second type of treatment assigned (i.e., the assignment of
the case to the Safe Streets Unit for follow-up investigation and counseling).

The data we used to examine personal and legal power were taken exclu-
sively from victim interviews for two reasons. First, we were interested in
how arrest and other police responses were related to the victims’ experi-
ences. Second, although there is a substantial research literature on the SARP
experiments, most studies, including the only one that examined pooled data
across all sites, analyzed suspect behavior. Because the victims were the con-
cern in this research, we examined how their personal and legal power can be
enhanced and thus used to prevent reoccurring or repeated domestic
violence.

Miller / AN ARRESTING EXPERIMENT 701

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

Interviews with the Dade County victims were conducted in Spanish or in
English, shortly after the initial domestic violence incident occurred (i.e., the
event that made the suspect and his partner eligible for the field experiment)
and 6 months later. The victims were paid $20 for each completed interview.
A total of 595 victims completed the first interview, but only 385 victims
completed the follow-up or second interview. The study’s attrition rate, simi-
lar to all the other SARP studies, has no verifiable explanation, although it
likely includes refusals from fearful women and the inability to locate women
who moved away from the suspects (Sherman, 1992). The analysis of the
data based on the second interviews, because of the high attrition rate, was
conducted for exploratory purposes only.

Measures

Victims, shortly after police intervention, reported to female interviewers
whether the domestic violence had continued. They also reported the type
and number of violent events that they experienced following the police call.
During their second or follow-up interviews, the victims reported the number
of physical assaults, threats, and property damage incidents that had
occurred. Based on victim responses during the first interview, we con-
structed a variable to indicate whether physical violence occurred subsequent
to a police call. Based on responses to the follow-up interview questions, we
counted the number of times a victim was hit, threatened, or experienced
property damage. We also constructed a summed scale to represent the total
number and type of incidents that victims reported over a period of 6 months.

Both interview schedules included items to measure each domestic part-
ner’s employment status, all sources and levels of income, levels of educa-
tional attainment, marital status, household composition, and whether the
victim and the suspect lived together. The initial interview included a single
question that asked victims how independent they are. Independence was
used to measure perceptions of personal power in this study. It was coded on a
5-point scale, in the direction of independence.

We constructed a measure of the victim’s income advantage that is based
only on categories of earned income: victim’s earned income divided by sus-
pect’s earned income. Values greater than 1.0 indicate that the victim had an
earned income advantage. Values less than 1.0 indicate that the victim had an
earned income disadvantage. Zero values indicate no earned income for one
or both domestic partners.

A single question asked how safe victims felt following police interven-
tion. Responses ranged from very unsafe (coded 1) to very safe (coded 4). The
victims reported whether they wanted the suspect arrested (yes or no) and

702 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

how satisfied they were, measured on a 4-point scale and coded in the direc-
tion of very safe, with what the police did in response to domestic violence
calls.

Six semantic-differential type items were used to measure legal power, or
the victim’s perceptions of how she was affected by the action that the police
took: (a) helpless or powerful, (b) out of control or in control, (c) afraid or
brave, (d) weak or strong, (e) discouraged or encouraged, and (f) hesitant or
determined. Respondents rated each item on a 7-point scale that was coded in
the direction of high levels of perceived power. Responses to the seven items
were summed to form a legal power scale. The Cronbach’s alpha (i.e., the
reliability measure for the summed scale) is .903.

The follow-up interview measured acts of violence that were perpetrated
by the suspect within 6 months after police intervention. A summed scale was
created to represent the total number of times the suspect threatened the vic-
tim, the number of assaults perpetrated, and the number of times the suspect
damaged the victim’s property. The Cronbach’s alpha for the reoccurring
violence scale is .804.

During the second or follow-up interviews, victims indicated how likely
or willing they were to call the police in the future if necessary. Willingness to
call was rated on a 3-point scale, in the direction of more likely to call. Vic-
tims rated the amount of stress they experienced in their relationships, coded
on a 0 to 4-point scale in the direction of increased stress. They told inter-
viewers whether the suspect recognized the wrongfulness of domestic vio-
lence. “Do not know” responses were coded 0.5, no responses were coded 0,
and yes responses were coded 1.0.

RESULTS

A Descriptive Profile

No segment of the adult population is immune to domestic violence. How-
ever, certain types of women, especially poor and minority women, are more
likely to be victimized and much more likely to be trapped within abusive
households (Hampton & Gelles, 1994; Mann, 1996; Richie, 1996; West,
1999). Moreover, police arrests and court actions affect a disproportionate
number of African Americans, relative to their representation in U.S. society
or their representation in the population of criminal offenders (Davis, 1997;
Gottfredson & Jarjoura, 1996; Hagan & Albonetti, 1982; Humphrey &
Fogarty, 1987; Jacobs & O’Brien, 1998; Klein, Petersilia, & Turner, 1990;
McCoy, 1997; Wortley, Macmillan, & Hagan, 1997). The Dade County arrest

Miller / AN ARRESTING EXPERIMENT 703

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

experiment, conducted in the urban area ranked seventh in the nation in
Latino population, appears to reflect the deeply institutionalized race dispar-
ity in the legal arena. The Dade County population, according to the 1990
U.S. census, was 20.5% African American, yet 42.6% of the suspects in the
Dade County experiment were African American. Compared to Anglo
women, African American women are much less likely to call the police to
arrest domestic violence suspects, or to use court procedures to stop the vio-
lence (Lee, Thompson, & Mechanic, 2002; Weis, 2001). However, African
American men, and their partners, were vastly overrepresented in the Dade
County experiment.

Approximately 21% of the couples in the experiment were Latino, 20%
were Anglo, and the remaining couples were mostly Asian American. Most
couples (79.4%) were married at the time, and 80% had at least one other per-
son, usually a child, living with them. Both the suspects and the victims
tended to have completed their formal education by earning a high school
diploma (71% of the suspects and 61% of the victims). At the follow-up inter-
view, 62% of the victims reported that they were employed and that 82% of
the suspects were employed.

Arrest and Reoccurring Domestic Violence

Table 1, based on the first victim interviews, shows the effect of arrest on a
binary-coded variable that indicates whether the victims experienced reoc-
curring physical violence shortly after police intervention. ANOVA was used

704 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

TABLE 1: Victim Reports of Violence Following Police Intervention, Suspect Assigned
to Control or Experimental Groupsa

More Violence Since
Police Intervention (Yes or No)

Treatment or Control Group (Actually Delivered) M SD n

No arrest, no Safe Streets (control group) 0.31 0.46 71
Arrest only 0.14 0.35 95
Safe Streets only 0.24 0.43 73
Both arrest and Safe Streets 0.18 0.39 100

Range 0-1
Overall M 0.21
Overall F (group differences, post hoc Tukey test) 2.761* (no arrest and no Safe

.Streets, and arrest only)

a. An ANOVA was run.

*p < .05.

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

to examine the statistical significance of differences in mean values across
groups, with a post hoc Tukey test to identify significant differences across
pairs of groups.

Overall, 21% of the victims reported that at least one episode of violence
followed the police intervention. We found that arrest, according to the Metro
Dade victims, had a moderate, short-term effect on reoccurring domestic vio-
lence. Of the victims in the control group (i.e., those whose partners were not
arrested or assigned to a Safe Streets Unit), 31% experienced subsequent acts
of violence shortly after the police call. Of the victims whose partners were
arrested (but not assigned to Safe Streets), 14% experienced reoccurring vio-
lence after the police call. The post hoc Tukey test showed that the only statis-
tically significant difference in reoccurring domestic violence was found
between the control and the arrest-only treatment group.

Personal power. An ordinary least squares regression model was specified
(see Table 2) to explain variance in the victim’s perception of independence
(personal power) as a function of suspect arrest, race, marital status and living
arrangements, and the victim’s social and economic resources. We found that
suspect arrest was not significantly related to the victim’s perception of per-
sonal power. This supports Mills’s (1998) distinction between the two types
of power that women can experience within their interpersonal relationships:
A legal response to violence was not related to perceptions of personal power.

Miller / AN ARRESTING EXPERIMENT 705

TABLE 2: Personal Power—Victim Perceptions of Independencea (n = 595)

Independent Variable b SE β t

Suspect arrested 0.018 0.116 .006 0.158
Anglo suspect –0.642 0.134 –.173*** –4.783
Married couple –0.540 0.146 –.137*** –3.705
Couple lives together –0.564 0.128 –.164*** –4.418
Suspect employed –0.790 0.155 –.196*** –5.104
Victim employed 0.847 0.133 .258*** 6.362
Victim earned income advantageb 0.289 0.121 .099* 2.383

Intercept 4.180 0.202 20.668
Adjusted R2 (F) .246*** (27.335)
M (SD) 3.200 (1.601)

a. A 5-point rating scale (1 = totally dependent, 5 = not dependent at all) was used, and an ordi-
nary least squares model was run.
b. Victim earned income advantage = victim’s earned income divided by suspect’s earned
income. Range = 0.05 to 4.80. Values less than 1.0 indicate victim’s disadvantage. Values greater
than 1.0 indicate victim’s advantage. A zero value indicates that either partner was unemployed
at the time of the police response.
*p < .05. ***p < .001.

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

A woman abused by an Anglo man in the Metro Dade County experiment,
compared to a woman abused by an African American or Latino man, experi-
enced less personal power. Richie’s (1996) gender entrapment theory offers a
counterintuitive explanation for this finding. Richie argued that a dual expo-
sure to racism and sexism makes African American women unusually vul-
nerable to domestic violence. The physical and emotional consequences of
violence within the home discourage women from reaching outside to social
control agencies that are presumed to be racist. Instead, many African Ameri-
can women are empowered by their relationships with friends and family to
control behaviors within their intimate relationships. African American
women are likely to “speak openly and directly about the violence in their
homes” (Weis, 2001, p. 156). Anglo women, however, are far more likely
than minority women to “deal silently with their ‘secret’” of domestic vio-
lence. They work to maintain the ideology of the “‘good’ white family life”
(Weis, 2001, p. 156). The contradiction, experiencing abuse while talking up
the “good husband,” can diminish or destroy perceptions of personal power
or independence.

We found that a victim’s perception of personal power was negatively
related to being married and to living with the suspect. The disadvantage of
marriage for some domestic violence victims has been documented empiri-
cally by family violence researchers, and it is explained by criminal opportu-
nity theory (McCloskey, 1996; Miller & Knudsen, 1999; Straus & Gelles,
1986). Domestic violence victims who are married to their offenders often
have little control over economic and social resources. Threats to leave a mar-
riage can result in the escalation of violence. Yet being married to, and living
with a domestic violence perpetrator, increases his opportunities to commit
reoccurring acts of violence.

In support of our research hypothesis, we found that employed victims,
compared to unemployed victims, perceived more personal power. The
greater the earned income advantage a victim had within her interpersonal
relationship, the more personal power she perceived. However, personal
power, contrary to our research hypothesis, was not related to whether the
victim experienced reoccurring violence following a police response. We
found (not reported here in table form) no significant empirical relationship
between a victim’s perception of independence and her report of domestic
violence following police intervention.

Legal power. We measured legal power with six items from the victim
interviews and used ANOVA to examine whether the arrest of a partner per se
was related to the degree of legal power perceived by the victims. All the

706 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

responses to the separate items were coded in the direction of increased
power, reflecting the degree to which victims felt more powerful, in control,
brave, strong, encouraged, and determined in response to the action taken by
the police.

The summed scale showed a high level of inter-item reliability
(Cronbach’s alpha = .903), but there was no significant difference in the legal
power scores across the victim-participant groups (not shown here in table
form). We noted, however, a distinctive pattern in the data. Victims whose
partners were arrested, compared to those whose partners were not arrested,
scored slightly higher on each item of the legal power scale. These “non-
findings” are potentially informative because they support Ford’s (1991)
argument that a criminal justice response can help a victim form an alliance
with a legal actor. The alliance may protect the victim from an escalation in
violence.

Table 3 shows the results of a regression model that was specified to
explain variance in legal power as a function of suspect arrest, the victim’s
preference for arrest, race, and the victim’s satisfaction with the police
response. We found that arrest per se was negatively related to perceptions of
legal power. It is quite possible that many women in the experiment wanted
the police to respond to their domestic violence problems but not to arrest
their partners (Mullings, 1997; Weis, 2001). This premise is supported
empirically. We found that if a victim wanted the police to arrest a suspect and
the police did arrest the suspect, the victim perceived a higher level of legal
power. The more satisfied she was with the police action that was taken,
whether or not the police action included arresting the suspect, the more legal
power she perceived. These findings clearly support Mills’s (1998, 1999)
argument that effective responses to domestic violence are those that reflect

Miller / AN ARRESTING EXPERIMENT 707

TABLE 3: Explaining Variance in Perceived Legal Powera (n = 588)

Independent Variable b SE β t

Suspect arrested –2.097 0.819 –.102** –2.560
Anglo suspect –3.319 0.866 –.141*** –3.832
Victim wanted and got suspect arrested 2.319 1.174 .077** 1.980
Victim satisfied with police action 3.444 0.291 .453*** 11.838

Intercept 18.026 1.109 16.250
Adjusted R2 (F) .213*** (40.792)
M (SD) 29.900 (10.148)

a. A 6-item scale was used (alpha = .903), and an ordinary least squares model was run.
**p < .01. ***p < .001.

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

the victim’s preferences and autonomy. A legal response that respects the
victim’s needs can have positive consequences. It can partner the victim
with a powerful social control agent and empower her to prevent reoccurring
violence.

The analysis shown in Table 4 partly supports Mills’s (1998, 1999) argu-
ment. We found that suspect arrest was positively related to a victim’s percep-
tion of safety. Feeling safe was not related to race or to the victim’s satisfac-
tion with the police action that was taken. However, perceptions of safety
were significantly related to perceptions of legal power. The more legal
power a victim perceived, the safer she felt following a domestic violence
incident. The analysis permits the inference that arrest can increase percep-
tions of safety, even for some victims who preferred the police to stop the vio-
lence without arresting the domestic violence suspect.

Six-Month Follow-Up Interviews

Due to the high attrition rate among the victim-participants in the Dade
County experiment, we analyzed the 6-month follow-up interviews as an
exploratory study. We drew inferences from our empirical findings only for
the purpose of encouraging discussion.

Table 5 shows that the three different types of reoccurring domestic vio-
lence that were measured by the follow-up victim interviews were not
affected by suspect arrest. We contend that on average, the victims in the
Dade County arrest experiment were unlikely to have experienced long-term
benefits as a consequence of suspect arrest. We also noticed that the standard
deviations, especially for batteries and threats (shown in Table 5), are sub-

708 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

TABLE 4: Victim Felt Safea (n = 588)

Independent Variable b SE β t

Suspect arrested 0.974 0.122 .327*** 7.964
Anglo suspect –0.007 0.129 –.002 –0.057
Victim wanted and got suspect arrested –0.221 0.173 –.051 –1.279
Legal power 0.044 0.006 .302*** 7.122
Victim satisfied with police action –0.028 0.048 .025 –0.572

Intercept 1.991 0.199 10.027
Adjusted R2 (F) .177*** (26.212)
M (SD) 3.83 (1.67)

a. A 4-point rating scale (1 = very unsafe, 4 = very safe) was used, and an ordinary least squares
model was run.
***p < .001.

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

709

TABLE 5: Six-Month Follow-Up Interviews, Victim Reports of Reoccurring Violencea

Number of Times Number of Times Number of Times
Hit or Battered Property Damaged Threatened

Treatment or Control Group (Actually Delivered) M SD n M SD n M SD n

No arrest, no Safe Streets control group 0.47 1.05 76 0.07 0.27 76 2.66 12.06 76
Arrest only 0.46 2.14 106 0.28 1.98 106 1.17 5.87 106
Safe Streets only 0.49 1.12 75 0.17 0.81 75 1.83 11.63 75
Both arrest and Safe Streets 0.26 0.88 115 0.06 0.27 115 0.59 2.48 115

Range 0-20 0-20 0-90
Overall M 0.41 0.15 1.42
Overall F 0.604 0.824 1.031

a. An ANOVA was run.

at S
A

G
E

P
ublications on D

ecem
ber 7, 2012

jiv.sagepub.com
D

ow
nloaded from

http://jiv.sagepub.com/

stantial. It is possible that for some victims, suspect arrest, without Safe
Streets follow up, resulted in an escalation of battery and threats. For other
victims, arrest could have prevented repeated acts or threats of violence.

In Table 6, a summed scale (Cronbach’s alpha = .804) that represents the
total number and types of reoccurring domestic violence is regressed on the
victim’s subjective experiences. We found that victims who, according to
their reports, experienced relatively high levels of stress in their marital or
intimate relationships also experienced higher levels of reoccurring violence.
Those who reported that the suspects realized the wrongfulness of domestic
violence reported less reoccurring violence.

Personal and Legal Power

Our research hypotheses, derived from Mills’s (1998, 1999) power model,
were partly supported by the analysis of the victim interviews that were con-
ducted as part of the Dade County arrest experiment. Women victimized by
Anglo suspects, ceteris paribus, perceived less personal power within their
intimate relationships and less legal power. Employed victims and those who
had an income advantage within their interpersonal relationships reported
relatively higher levels of personal power. In principle, personal power gives
victims a tool or an instrument to prevent reoccurring domestic violence.
Empirically, we could not, however, confirm the expected relationship
between the victim’s personal power and the suspect’s desistance of domestic
violence following a police response.

710 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

TABLE 6: Victim Reports of Domestic Violence Following Police Intervention, Sum of
Number of Times Hit, Threatened, and Property Damaged (alpha = .804)
(n = 345)

Independent Variable b SE β t

Victim’s perception of relationship stress 0.363 0.128 .157** 2.838
Victim thinks she is more likely to call

police in future 0.160 0.212 .039 0.754
Suspect realizes wrongfulness of

domestic violence –1.312 0.321 –.229*** –4.082

Intercept 0.414 0.701
Adjusted R2 (F) .102 (12.948)***
M (SD) 0.834 (2.619)

a. An ordinary least squares model was run.
**p < .01. *** p < .001.

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

A police response to domestic violence, including suspect arrest, can
increase the victim’s legal power that can be used to prevent reoccurring vio-
lence. The more satisfied a victim was with the police action that was taken,
the more legal power she perceived. Victims who experienced high levels of
legal power felt more safe as they anticipated and controlled future social
interactions with their partners.

Six months following police intervention, it was the level of stress within a
relationship and the victim’s perception that the suspect recognized the
wrongfulness of domestic violence that were related to the probability of
reoccurring violence. Based on these findings, we posit that the most reason-
able criminal justice and social service responses to domestic violence are
those that consider the victim’s needs by taking into account her subjective
experiences, her cultural and social resources, and her personal and legal
resources. In addition, the most effective responses are likely to be those that
convincingly demonstrate, to the suspect, the wrongfulness of domestic
violence.

DISCUSSION

Method Issues

The limitations of this study are clear. The research participants, all
women, were in heterosexual relationships and experienced at least one inci-
dent of misdemeanor domestic violence that was brought to the attention of
the police. The victim-participants in the arrest experiment do not represent
victims in Dade County, or domestic violence victims in other areas of the
United States. Most serious are the disadvantages imposed by the short-term
(6 month) victim follow-up period and by the unacceptably high attrition rate
among the research participants.

This study also makes clear the advantages of an experimental field
design. We pose two crucial questions that all experiments should ask and
answer: Did the Dade County arrest experiment cause harm to domestic vio-
lence victims? We found no evidence that victims faced an increased likeli-
hood of reoccurring violence as a consequence of the arrest experiment. Did
the failure to arrest those randomly assigned to a control group cause victim
harm? We conclude, ironically, that it did not. Arrest may have had a statisti-
cally significant albeit weak effect on the probability of reoccurring domestic
violence.

Miller / AN ARRESTING EXPERIMENT 711

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

Responses to Domestic Violence

Our analysis of the Metro Dade victim data contributes to the literature on
social and legal responses to domestic violence. We conclude that personal
and legal power are subjectively experienced perceptions that can be effec-
tive resources for domestic violence victims. Legal actors can form partner-
ships with victims by recognizing that each person is unique and faces cul-
tural, economic, family, and emotional circumstances that can increase or
decrease the probability of reoccurring violence. Partnerships and alliances
empower victims. They are, however, precluded by police or court actions
that fail to consider the unique victim’s characteristics and needs.

Perceptions of personal power can be reaffirmed by legal power. Together,
personal and legal power can be used to influence and control the suspect’s
behaviors, as they simultaneously assure the victim’s perceptions of safety.

This research, because it is based on an arrest experiment that included an
extremely disproportionate representation of African American victims,
accentuates the need for domestic violence programs to appeal to our African
American communities. Police arrests, safe shelters, and prosecution pro-
grams have been the preferred solutions for domestic violence problems in
the United States since the mid-1970s. However, many African American
women remain unwilling to turn to safe shelters because they are not “cultur-
ally friendly” (Nelson, 2002, p. 2). In other research (Miller, 2002), we found
that African American victims were compelled to move from a safe shelter to
a homeless shelter to avoid assault by Anglo clients within the domestic vio-
lence shelter.

Other victims refuse to call the police to avoid turning their partners over
to a criminal justice system that, they perceive, discriminates against African
Americans (Nelson, 2002, p. 4). African American victims can be empow-
ered by police and other sociolegal actors who recognize the circumstances
that the individual African American victim and her community encounter.
Oliver (2000) illustrated the possibilities. He recognized the limits of typical
domestic violence programs that are based on what he called a “one size fits
all” model, and he urged the development of prevention and intervention pro-
grams that are based on African American popular culture. He cited success-
ful programs that focus on culture-specific radio campaigns, gospel music,
and African American icons in public service announcements. Some advo-
cates may argue that only our urban areas with the most diverse populations
and the healthiest fiscal conditions can afford the culturally diverse programs
that are needed to respond to the various types of domestic violence victims
in the United States. We agree with Oliver and with Richie (2000) who

712 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

reminded advocates that “the assumed race and class neutrality of gender
violence led to the erasure of low-income women and women of color from
the dominant view” (p. 1135). No city and no intervention program can
afford to ignore all the violence and all the victims.

Future Research and Domestic Violence Perspectives

Social science theories of domestic violence tend to explain the reoccur-
rence or desistance of battery and threats of violence (Miller & Knudsen,
1999). Many feminist perspectives examine the consequences of patriarchy
for women in general and for specific women within their intimate relation-
ships (Dobash, Dobash, Cavanagh, & Lewis, 2000). Other feminists plait
race and class into their explanations (Richie, 1996; Weis, 2001). We offer
three modest suggestions for the continued development of perspectives that
focus on what women can do to prevent and stop domestic violence.

First, we argue that domestic violence theory and research should con-
tinue to focus on explaining the subjective experiences of women. Emotions
and perceptions, such as empowerment, stress, and feeling safe, can have
important effects for women who face the risk of domestic violence. Emo-
tions and perceptions are central because outside actors, the police or
extended family members, do not share a bedroom with a potential abuser.

Second, we argue that a woman’s culture, her resources, and her connec-
tions to legal actors can enhance perceptions of personal and legal power.
Cultural and social resources can empower women to talk and disclose
shared problems, thus insulating women from the dangers of isolation.
Shared accounts can protect individual women. Financial resources, espe-
cially earned income, can empower women by making them agents of social
control within an intimate relationship. Legal alliances can enhance percep-
tions of safety and trust.

Third, we take the feminist position that the only reasonable explanation
of domestic violence is one that considers simultaneously the unique person
and the intersection of race, gender, and class in U.S. society. We contend that
it is absolutely unacceptable for any woman to be subjected to the injuries of
domestic violence. Concomitantly, we contend that an explanation of domes-
tic violence that fails to address race and class differences is insufficient.

Although “every woman” can be a victim of domestic violence, according
to the slogan, I realize full well that I sit comfortably in my office to write
about a problem that too many women, often poor, homeless, and minority
women, will not get the opportunity to avoid.

Miller / AN ARRESTING EXPERIMENT 713

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

REFERENCES

Berk, R. A. (1993a). Reply: Policy correctness in the ASR. American Sociological Review, 58,
889-890.

Berk, R. A. (1993b). What the scientific evidence shows: On average, we can do no better than
arrest. In R. J. Gelles & D. R. Loseke (Eds.), Current controversies on family violence
(pp. 323-336). Newbury Park, CA: Sage.

Berk, R. A., Campbell, A., Klap, R., & Western, B. (1992). The deterrent effect of arrest in inci-
dents of domestic violence: A Bayesian analysis of four field experiments. American Socio-
logical Review, 57, 698-708.

Davis, A. J. (1997). Race, cops, and traffic stops. University of Miami Law Review, 51, 425-443.
Davis, R. C., & Taylor, B. G. (1997). A proactive response to family violence: The results of a

randomized experiment. Criminology, 35, 307-333.
Dobash, R. P., Dobash, R. E., Cavanagh, K., & Lewis, R. (2000). Confronting violent men. In

J. Hanmer & C. Itzin (Eds.) (with S. Quaid & D. Wigglesworth), Home truths about domestic
violence: Feminist influences on policy and practice—A reader (pp. 289-309). London:
Routledge.

Dunford, F., Huizinga, D., & Elliott. D. S. (1991). The role of arrest in domestic assault. The
Omaha experiment. Criminology, 28, 183-206.

Ford, D. A. (1991). Preventing and provoking wife battery through criminal sanctions: A look at
the risks. In D. D. Knudsen & J. L. Miller (Eds.), Abused and battered: Social and legal
responses to family violence (pp 191-209). New York: Aldine.

Ford, D. A., & Regoli, M. J. (1993). The criminal prosecution of wife assaults: Process, prob-
lems, and effects. In N. Z. Hilton (Ed.), Legal responses to wife assault: Current trends and
evaluation (pp. 127-164). Newbury Park, CA: Sage.

Ford, D. A., & Regoli, M. J. (1998). The Indianapolis domestic violence prosecution experi-
ment. In National Institute of Justice and American Bar Association (Eds.), Legal interven-
tions in family violence: Research findings and policy implications (pp. 62-64). Washington,
DC: U.S. Department of Justice.

Garner, J., Fagan, J., & Maxwell, C. (1995). Published findings from the Spouse Assault Repli-
cation Program: A critical review. Journal of Quantitative Criminology, 11, 3-28.

Gelles, R. J. (1993). Constraints against family violence: How well do they work? American
Behavioral Scientist, 36, 575-586.

Gottfredson, S. D., & Jarjoura, G. R. (1996). Race, gender, and guidelines-based decision mak-
ing. Journal of Research in Crime and Delinquency, 33, 49-69.

Hagan, J., & Albonetti, C. (1982). Race, class, and the perception of criminal injustice in Amer-
ica. American Journal of Sociology, 88, 329-355.

Hampton, R. L., & Gelles, R. J. (1994). Violence toward Black women in a nationally represen-
tative sample of Black families. Journal of Comparative Family Studies, 25, 105-119.

Hirschel, J. D., Hutchison, I. W., III, & Dean, C. W. (1992). The failure of arrest to deter spouse
abuse. Journal of Research in Crime & Delinquency, 29, 7-33.

Humphrey, J. A., & Fogarty, T. J. (1987). Race and plea bargained outcomes: A research note.
Social Forces, 66, 176-182.

Jacobs, M. S., & O’Brien, R. M. (1998). The determinants of deadly force: A structural analysis
of police violence. American Journal of Sociology, 103(4), 837-862.

Jasinski, J. L. (2001a). Physical violence among Anglo, African American, and Hispanic cou-
ples: Ethnic differences in persistence and cessation. Violence and Victims, 16, 479-490.

714 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

Jasinski, J. L. (2001b). Pregnancy and violence against women: An analysis of longitudinal data.
Journal of Interpersonal Violence, 16, 712-733.

Jasinski, J. L. (2001c). Theoretical explanations for violence against women. In C. M. Renzetti,
J. L. Edelson, & R. K. Bergen (Eds.), Sourcebook on violence against women (pp. 5-22).
Thousand Oaks, CA: Sage.

Johnson, I. M. (1992). Economic, situational, and psychological correlates of the decision-
making process of battered women. Families in Society: The Journal of Contemporary
Human Services, 73, 168-176.

Klein, S., Petersilia, J., & Turner, S. (1990). Race and imprisonment decisions in California. Sci-
ence, 247, 812-816.

Lee, R. K., Thompson, V. L., & Mechanic, M. B. (2002). Intimate partner violence and women of
color: A call for innovations. American Journal of Public Health, 92, 530-534.

Lempert, R. (1989). Humility is a virtue: On the publication of policy relevant research. Law and
Society Review, 23, 145-161.

Mann, C. R. (1996). When women kill. Albany: State University of New York Press.
Maxwell, C., Garner, J., & Fagan, J. (1999). The impact of arrest on domestic violence: Results

from five policy experiments. National Institute of Justice Journal, 241, 27-28.
Maxwell, C., Garner, J., & Fagan, J. (2001). The effects of arrest on intimate partner violence:

New evidence from the Spouse Assault Replication Program. Research in brief. Washington,
DC: U.S. Department of Justice.

McCloskey, L. A. (1996). Socioeconomic and coercive power within the family. Gender & Soci-
ety, 10, 449-463.

McCoy, C. (1997). Sentencing and the underclass. Law & Society Review, 31, 589-612.
Miller, J. (2002). Remedial work: Accounts of violent events and interpersonal relationships.

Unpublished manuscript.
Miller, J. L., & Knudsen, D. (with Copenhaver, S.) (1999). Family abuse and violence. In M. B.

Sussman, S. K. Steinmetz, & G. W. Peterson (Eds.), Handbook of marriage and the family
(pp. 705-756). New York: Plenum.

Mills, L. G. (1998). Mandatory arrest and prosecution policies for domestic violence: A critical
literature review and the case for more research to test victim empowerment approaches.
Criminal Justice and Behavior, 25, 306-319.

Mills, L. G. (1999). Commentary: Killing her softly: Intimate abuse and the violence of state
interventions. Harvard Law Review, 113, 550-611.

Mullings, L. (1997). On our own terms: Race, class and gender in the lives of African American
women. New York: Routledge.

Nelson, L. I. (2002). Community solutions to domestic violence must address cultural roots and
beliefs. Assembling the Pieces: An African American Perspective on Community and Family
Violence, 3, 2-5.

Oliver, W. (2000). Preventing domestic violence in the African American community: The ratio-
nale for popular culture interventions. Violence Against Women, 6, 533-549.

Pate, A. M., & Hamilton, E. E. (1992). Formal and informal deterrents to domestic violence: The
Dade County spouse assault experiment. American Sociological Review, 57, 691-697.

Pate, A. M., Hamilton, E. E., & Annan, S. (1994). Spouse abuse replication project in Metro-
Dade County Florida, 1987-1989. Washington, DC: Police Foundation, 1991 (Producer).
Ann Arbor, MI: Inter-University Consortium for Political and Social Research (Distributor).

Richie, B. E. (1996). Compelled to crime: The gender entrapment of battered Black women. New
York: Routledge.

Richie, B. E. (2000). A Black feminist reflection on the antiviolence movement. Signs: Journal
of Women in Culture and Society, 25, 1113-1137.

Miller / AN ARRESTING EXPERIMENT 715

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/

Sherman, L. W. (1992). Policing domestic violence: Experiments and dilemmas. New York: Free
Press.

Sherman, L. W. (1993). Deviance, deterrence, and irrelevance: A theory of criminal sanction.
Journal of Research in Crime & Delinquency, 30, 445-473.

Sherman, L. W., & Berk, R. A. (1984). The specific deterrent effects of arrest for domestic
assault. American Sociological Review, 49, 261-272.

Sherman, L. W., Schmidt, J. D., Rogan, D. P., Gartin, P. R., Cohn, E. G., Collins, D. J., et al.
(1991). From initial deterrence to long-term escalation: Short-custody arrest for poverty
ghetto domestic violence. Criminology, 29, 821-850.

Stephens, B. J., & Sinden, P. G. (2000). Victims’ voices: Domestic assault victims’ perceptions
of police demeanor. Journal of Interpersonal Violence, 15, 534-547.

Straus, M. A., & Gelles, R. A. (1986). Societal change in family violence from 1975 to 1985 as
revealed by two national surveys. Journal of Marriage and the Family, 48, 465-479.

Teichman, M., & Teichman, Y. (1989). Violence in the family: An analysis in terms of interper-
sonal resource-exchange. Journal of Family Violence, 4, 128-139.

Toby, J. (1957). Social disorganization and stake in conformity: Complementary factors in the
predatory behavior of hoodlums. Journal of Criminal Law, Criminology and Police Science,
48, 12-17.

Weis, L. (2001). Race, gender, and critique: African-American women, White women, and
domestic violence in the 1980s and 1990s. Signs: Journal of Culture and Society, 27, 139-
169.

West, T. C. (1999). Wounds of the spirit: Black women, violence and the resistance ethic. New
York: New York University Press.

Wortley, S., Macmillan, R., & Hagan, J. (1997). Just des(s)erts? The racial polarization of per-
ceptions of criminal injustice. Law & Society Review, 31, 637-676.

JoAnn Miller is an associate professor in the Department of Sociology and Anthropology
at Purdue University. Her research and interests focus on social problems, social
inequalities, and interpersonal violence. She and Robert Perrucci are the editors of Con-
temporary Sociology.

716 JOURNAL OF INTERPERSONAL VIOLENCE / July 2003

at SAGE Publications on December 7, 2012jiv.sagepub.comDownloaded from

http://jiv.sagepub.com/