CCJ5702 Individual Work Week 4 (Stephen Scott)

10.1177/1088767905285439Homicide StudiesAddington / Murder Data and Clearance Predictors

Using National Incident-Based Reporting System Murder Data to Evaluate Clearance Predictors A Research Note

Lynn A. Addington American University, Washington, DC

One limitation with the Uniform Crime Reporting (UCR) Program’s Supplementary Homicide Report (SHR) is its failure to indicate whether a particular murder has been cleared. As a result, researchers using the SHR must rely on proxy measures to study clearance at the national level. Currently, the UCR Program is undergoing a large-scale conversion from its traditional summary system and SHR to the National Incident-Based Reporting System (NIBRS). One benefit of NIBRS is that it enables law enforcement agencies to report incident-level clearance information. Although NIBRS provides a new and important source of clearance information, researchers have largely ignored these data. The present study provides an initial examination of the NIBRS murder clearance data. Specifically, these data are used both to evaluate clearance predictors and to assess the validity of the proxy clearance measures previously used with the SHR.

Keywords: National Incident-Based Reporting System; murder data; clearance

Murder clearance rates are important criminal justice measures for researchersand policy makers.1 Clearance rates provide a readily available—if limited and problematic (Riedel & Jarvis, 1999)—measure of police performance by describing the degree to which crimes are solved. Currently, this measure is under particular scru- tiny because the percentage of cleared murders has decreased steadily from 93% in 1960 to 65% in 1995 (Riedel, 1999) and 62% in 2003 (Federal Bureau of Investigation [FBI], 2004a). Despite the interest in clearance rates, relatively little is known about these crime statistics and only a handful of studies have examined factors related to murder clearance. One reason for this dearth of knowledge is a lack of available data. Although homicide researchers typically use data from the Uniform Crime Reporting (UCR) and its Supplementary Homicide Report (SHR), both of these resources have significant limitations for studying clearance issues. Although the summary UCR sys- tem collects clearance information, these aggregate-level data do not identify which

140

Homicide Studies Volume 10 Number 2

May 2006 140-152 © 2006 Sage Publications

10.1177/1088767905285439 http://hs.sagepub.com

hosted at http://online.sagepub.com

Author’s Note: An earlier version of this study was presented at the 2004 Homicide Research Working Group Annual Summer Workshop in Ann Arbor, MI. I thank John Jarvis at the Federal Bureau of Investiga- tion for sharing his extensive knowledge of National Incident-Based Reporting System data.

particular murder was cleared. While the SHR gathers incident-level details, no arrest or clearance data are collected.

Currently, the UCR Program is in the process of a substantial conversion from its traditional summary system to the National Incident-Based Reporting System (NIBRS). Unlike the summary system and its SHR, NIBRS provides incident-level details, including clearance information for murders as well as more than 40 other crimes (FBI, 1992). Although NIBRS provides a new and important source of clear- ance information, researchers have largely ignored these data (but see Addington, 2005; Chilton & Jarvis, 1999). This oversight is unfortunate because one of the goals in redesigning the UCR Program was to improve clearance information to allow for the exploration of these statistics (Poggio, Kennedy, Chaiken, & Carlson, 1985). The present study is the first to examine these data and assess their utility for studying murder clearance.

To explore the NIBRS clearance data, the remainder of this research note is orga- nized in the following manner. The first section summarizes previous research on homicide clearance. The NIBRS data and methods used in this study are described. Then the findings obtained are presented and discussed.

Background

Little attention has been devoted to studying murder clearance. This section reviews the existing research by focusing on (a) the characteristics associated with cleared murders and (b) the ways in which previous researchers have measured clear- ance, particularly the proxy measures used with SHR data.

Characteristics Related to Clearance

Incident characteristics as well as police response affect the likelihood a murder is cleared. The following discussion concentrates on the incident-related characteristics associated with clearance because this information is collected by NIBRS and is the focus of the present study.2 These factors include victim demographics, victim- offender relationship, weapons used, and location of the incident.

Characteristics related to murder victims, such as their demographics and relation- ship with the offender, are associated with clearance. Researchers have examined vic- tim demographics to test the hypothesis that police use extralegal factors in allocating the time and effort spent investigating murders (e.g., Puckett & Lundman, 2003). Under this hypothesis, more attention would be given to solving murders involving affluent or sympathetic victims (such as victims who are women, children, or White), and these cases would have higher clearance rates. Findings from existing studies pro- vide mixed support for this idea of victim preferencing by police. Murders involving young victims are more likely to be cleared than those involving older victims (Litwin, 2004; Puckett & Lundman, 2003; Riedel & Rinehart, 1996), and those involving females are more likely to be cleared than murders involving male victims (Regoeczi, Kennedy, & Silverman, 2000). However, murders involving non-White victims are

Addington / Murder Data and Clearance Predictors 141

more likely or just as likely to be cleared as those involving White victims (Litwin, 2004; Puckett & Lundman, 2003; Regoeczi et al., 2000). Researchers also have focused on the victim-offender relationship in studying clearance. Victim-offender relationship is of interest because an increase in murders among strangers is one ex- planation given for the rise in uncleared murders during the past few decades (Riedel, 1998; Wellford & Cronin, 1999). This hypothesis has some support as family-related murders are more likely to be cleared than nonfamily homicides, particularly stranger murders (Riedel, 1998).

Homicide clearance studies also have examined factors that may affect the evi- dence available for investigators, such as weapons used and location of the crime. As part of their assessment of clearance predictors, Puckett and Lundman (2003) com- pared weapons that were more likely to involve victim-offender contact and leave evi- dence (such as knives, personal weapons, and blunt objects) with those less likely to involve contact (such as firearms). They found murders involving knives and other “contact weapons” more likely to be cleared than those involving firearms. Where a murder occurs is also related to clearance (Litwin, 2004; Wellford & Cronin, 1999). Murders that occur in a private location such as a home are more likely to be cleared than those that occur in a public area such as a street (Wellford & Cronin, 1999). One explanation for this finding is that private locations protect crime scene evidence, which can assist detectives in identifying a suspect.

Clearance Measures

Previous studies have had to rely on clearance measures that compromise either coverage or accuracy. Researchers desiring accurate clearance measures must col- lect these data from local law enforcement agencies.3 The time and expense of such investigator-initiated data collection efforts limit these studies to a single city (Litwin, 2004; Puckett & Lundman, 2003) or a small number of jurisdictions (Wellford & Cronin, 1999). Alternatively, researchers interested in broader coverage must rely on UCR and SHR data, which lack clearance information linked to particular murders. Researchers must generate proxy clearance measures to use the SHR data (e.g., Regoeczi et al., 2000; Riedel & Rinehart, 1996).

Two examples of proxy clearance measures developed for the SHR are those used by Regoeczi and her colleagues (2000) and Riedel and Rinehart (1996). Regoeczi and her colleagues developed a proxy measure to analyze 17 years of SHR data. Uncleared murders are defined as those cases involving “unknown offenders” and were identi- fied using the SHR situation codes of single victim/unknown offender and multiple vic- tim/unknown offender. These situations occur when nothing is known about the offender, not even the number of offenders involved. Reidel and Rinehart used a slightly different proxy measure for their analysis of 5 years of SHR-based data for Chicago. Here, cases are considered uncleared if no demographic characteristics (age, race, and sex) are known about the offender. This definition of uncleared is broader than that used by Regoeczi and her colleagues because it includes all the unknown offender cases as well as cases in which the number of offenders is known, but no

142 Homicide Studies

Addington / Murder Data and Clearance Predictors 143

demographic details were reported. Such a situation could occur if a witness saw two individuals running from the crime scene but the suspects were wearing ski masks so none of their features were observed. Both proxy measures are likely to overestimate clearance because knowing some information about an offender does not guarantee that the case will be cleared. This overestimation would not be a problem if the same patterns were observed for both proxy and actual clearance measures. The validity of these measures, although, is unknown and is one focus of this study.

Research Questions

The present study seeks to explore the utility of NIBRS data for examining clear- ance issues and focuses on two preliminary, but currently unexamined, questions. The first question is “What incident-level characteristics predict murder clearance based on actual clearance measures from NIBRS?” Answering this first question is impor- tant considering that no prior research has used NIBRS data to study clearance predic- tors. In addition, no clearance study has been conducted at the national level using actual clearance measures.4 Prior studies using actual clearance measures have been limited to, at most, a handful of cities. A second and somewhat related question is “How valid are the proxy clearance measures used in connection with the SHR as compared with the actual measure from NIBRS?” The answer to this second question remains relevant despite the introduction of NIBRS. During the conversion to NIBRS, national homicide incident data continue to be reported in SHR format (FBI, 2004a). The SHR also provides a long time series of homicide data. If a proxy measure is valid, research- ers can use that proxy with greater confidence to study clearance issues using the SHR.

Data and Method

Data

This study uses data from the 2001 NIBRS (FBI, 2004b). These data constitute one of the most recently available years of public use data and are available through the National Archive of Criminal Justice Data. The cases analyzed are all murders and nonnegligent manslaughters. Both offenses are referred to as murder for shorthand. The unit of analysis is murder victims. In 2001, 1,958 murder victims were reported in NIBRS.5

NIBRS murder data are similar to SHR data.6 Both NIBRS and SHR data are inci- dent based and provide victim and offender demographics as well as details such as victim-offender relationship, circumstance, and weapon. NIBRS provides additional information the SHR does not. Of relevance for this study is that NIBRS collects infor- mation about the murder from its initial report through its clearance7 as well as addi- tional incident details such as location of the crime.

One caveat in analyzing NIBRS data is its limited coverage. NIBRS is a substantial departure in crime data collection for law enforcement agencies and requires a lengthy certification process. As a result, the conversion to NIBRS has been gradual. NIBRS

agencies covered only 17% of the U.S. population in 2001 (Bureau of Justice Statistics [BJS], 2004). In 2001, only 21 states were NIBRS certified.8 Within these 21 states, not all agencies submit data in NIBRS format. Only 7 states fully report in NIBRS; the other 14 states have less than full participation in NIBRS (BJS, 2004). Law enforce- ment agencies that participate in NIBRS tend to represent smaller population areas. In 2001, no agency covering a population of more than 1 million participated in NIBRS.9

These factors result in a large case selection bias, which poses two important limita- tions for this study. First, because participation in NIBRS is voluntary, NIBRS states and law enforcement agencies do not constitute a representative sample of U.S. law enforcement agencies or states (Addington, Loftin, & McDowall, 2001). Second, only a fraction of murders are reported to NIBRS. In 2001, only 1,958 of the 15,980 U.S. murders (or 12%) were reported to the UCR in NIBRS format.

This nonrepresentativeness of NIBRS suggests exercising caution when interpret- ing the results and generalizing beyond the NIBRS-participating agencies included in this study. It is possible that clearance patterns may vary across jurisdictions, espe- cially for those police agencies covering larger population areas. Given this caveat, understanding the capabilities of NIBRS for studying clearance issues is increasingly important as more states and agencies convert from summary and SHR reporting to NIBRS.

Variables

Clearance measures. Three clearance measures are used to examine how a direct measure of clearance compares with two proxy measures. Table 1 provides frequen- cies and coding information for all variables used in this study. The first clearance measure is a direct measure from NIBRS and is referred to as actual clearance. Mur- ders are classified as either cleared or uncleared. Cleared murders are those identified as either cleared by arrest or exceptionally cleared. A murder is cleared by arrest if at least one offender is arrested, charged, and turned over to the court for prosecution in connection with the murder (FBI, 2004a). Exceptionally cleared cases occur when a suspect is identified but events beyond the law enforcement agency’s control prevent an arrest because of the death of the offender, prosecution being declined, or extradi- tion being denied (FBI, 1992). For NIBRS murders occurring in 2001, 66% were cleared, and the vast majority of these cases were cleared by arrest (see appendix).10

Clearance is also measured by replicating two proxies previously used with SHR data. Here NIBRS data are used to recreate these measures. This replication is possible because NIBRS collects information comparable to the SHR measures used to gener- ate the proxies. The first proxy is referred to as the unknown offender measure and is based on the clearance measure used by Recoegczi and her colleagues (2000). The unknown offender variable defines uncleared murders as those in which no informa- tion is known about the offender. NIBRS identifies these offenders using a specific code.11 In this study, murders with the offender so identified are coded as uncleared, and the rest are considered cleared. As shown on Table 1, a higher percentage of mur- ders are identified as eared using the unknown offender proxy as compared to the

144 Homicide Studies

Addington / Murder Data and Clearance Predictors 145

actual clearance measure (88% vs. 66%). The second proxy measure is referred to as the unknown offender demographics measure, which is based on the Reidel and Rinehart (1996) proxy. This measure defines uncleared cases as those included in the unknown offender proxy as well as those where none of the three demographic char- acteristics (age, race, sex) is known about the offender. NIBRS identifies whether this information is present. This study used the offender demographic information to create a marker to identify those cases in which all three demographic characteristics were unknown. The unknown offender demographic measure identifies slightly fewer cases as cleared than the unknown offender measure (81% vs. 88%), but the unknown offender demographic proxy still counts more cases as cleared than does the direct clearance measure (Table 1).

Table 1 Frequencies and Metrics for Murder Clearance Models, 2001 NIBRS

Variable Code Frequency %

Clearance measure

Actual 1 = Cleared 1,287 66

0 = Uncleared 671 34 Unknown offender 1 = Cleared 1,725 88

0 = Uncleared 233 12 Unknown offender demographics 1 = Cleared 1,589 81

0 = Uncleared 369 19 Clearance predictors

Victim sex 1 = Female 584 30 0 = Male 1,359 70 Missing 15 < 1

Victim race 1 = White 1,026 52 0 = Non-White 883 45 Missing 49 3

Victim age Less than 12 111 6 13 to 59 1,587 81 Older than 60 175 9 Missing 85 4

Weapon used Gun 1,141 58 Knife 248 13 Contact weapon 305 16 Other weapon 111 6 Missing 153 8

Home location 1 = Home 1,087 56 0 = Other location 871 45

More than one victim 1 = More than one victim 476 24 0 = Only one victim 1,482 76

Note: Percentages may not total 100% due to rounding. N = 1,958.

Clearance predictors. The variables used to predict clearance are based on the pre- vious studies summarized in the Background section and include victim demograph- ics and incident details. The victim demographic variables include age, race, and sex. Based on prior clearance studies, age is categorized into three groups to compare chil- dren and adults (including teenagers) with those victims older than 65. For race, White and non-White victims are compared due to the small number of minority victims who are not African American (e.g., Asians and Native Americans). Sex uses males as the comparison group.

The incident details include weapon used, location, and more than one victim.12

Weapon used is categorized into four groups: firearm (both long guns and handguns), knife, contact weapons (asphyxiation, blunt objects, and personal weapons), and other (such as poison or fire). The firearm category is used as the comparison group based on Puckett and Lundman’s (2003) work. Although NIBRS allows for reports of up to three weapons per each offense in the incident, the decision was made to count only the first weapon reported. This decision rule greatly simplifies the analysis and includes the vast majority of weapons, because 85% of the murders involved only one weapon. Location is a binary measure that compares murders occurring in a residence with those occurring in other locations. More than one victim controls for multiple vic- tim cases because these cases are slightly more likely to be cleared (see the appendix). Here multiple victim cases refer to homicide incidents involving more than one crime victim, not necessarily more than one homicide victim.

Analyses Conducted

Because the outcome measure is a dichotomy (cleared vs. uncleared murders), binary logistic regression models are used to model predictors of clearance. Missing data for the clearance predictor variables are handled using complete case analysis wherein cases with missing data are dropped from the analysis. Most of these vari- ables have only a small amount of missing data (less than 5%; Table 1). As such, com- plete case analysis is an acceptable way to handle this issue (Allison, 2002).

Findings

Table 2 reports the logistic regression models and the adjusted odds ratios for the three clearance measures. The actual clearance measure column indicates that four incident-level characteristics affect clearance: age of the victim, weapon used, loca- tion of the murder, and number of victims. Specifically, murders of young victims are more likely to be cleared than those involving older victims. Cases in which a knife or contact weapon is used are more likely to be cleared than cases with a firearm. Mur- ders occurring in a residence are more likely to be cleared than those occurring outside the home. Finally, murder incidents involving more than one victim are more likely to be cleared as compared to cases involving only one victim.

Looking across the columns in Table 2, it is observed that a few different conclu- sions would be drawn if either of the proxy models were used rather than the actual

146 Homicide Studies

147

T ab

le 2

L og

is ti

c R

eg re

ss io

n M

od el

E xp

la in

in g

C le

ar ed

M u

rd er

s U

si n

g A

ct u

al a

n d

P ro

xy C

le ar

an ce

M ea

su re

s, 2

00 1

N IB

R S

A ct

ua l

C le

ar an

ce M

ea su

re

P ro

xy C

le ar

an ce

M ea

su re

s

U nk

no w

n O

ff en

de r

U nk

no w

n O

ff en

de r

D em

og ra

ph ic

s

A dj

us te

d A

dj us

te d

A dj

us te

d O

dd s

R at

io s

S E

z S

co re

p V

al ue

O dd

s R

at io

s S E

z S

co re

p V

al ue

O dd

s R

at io

s S E

z S

co re

p V

al ue

F em

al e

vi ct

im 1.

1 0.

1 1.

0 .3

1. 4

0. 3

1. 5

.1 1.

5 0.

3 2.

3 .0

2* W

hi te

v ic

ti m

1. 1

0. 1

1. 1

.3 1.

8 0.

3 3.

4 .0

01 *

1. 5

0. 2

3. 0

.0 03

* V

ic ti

m a

ge 1

2 or

y ou

ng er

a 2.

2 0.

8 2.

2 .0

3* 1.

5 1.

1 0.

5 .6

4. 0

2. 6

2. 1

.0 3*

V ic

ti m

a ge

1 3

to 5

9a 0.

9 0.

2 –0

.3 .8

0. 5

0. 2

–1 .5

.1 0.

8 0.

2 –0

.7 .5

K ni

fe b

1. 8

0. 3

3. 5

.0 01

* 2.

1 0.

7 2.

3 .0

2* 1.

6 0.

4 2.

0 .0

4* C

on ta

ct w

ea po

nb 1.

4 0.

2 2.

3 .0

2* 1.

3 0.

4 0.

9 .4

1. 2

0. 3

0. 8

.4 O

th er

w ea

po nb

1. 1

0. 3

0. 5

.6 0.

9 0.

3 –0

.2 .8

1. 0

0. 3

0. 1

.9 H

om e

lo ca

ti on

1. 4

0. 2

3. 3

.0 01

* 1.

4 0.

3 2.

1 .0

4* 1.

5 0.

2 2.

7 .0

07 *

M or

e th

an o

ne v

ic ti

m 1.

7 0.

2 4.

2 <

.0 01

* 2.

4 0.

5 3.

8 <

.0 01

* 1.

9 0.

3 3.

6 <

.0 01

*

N ot

e: N

= 1,

71 0.

a. C

om pa

ri so

n gr

ou p

is vi

ct im

s o ld

er t

h a n 6

0. b.

C om

pa ri

so n

gr ou

p is

fi re

a rm

s. *p

< .0

5.

148 Homicide Studies

clearance measure. These differences can be viewed in terms of (a) statistically signif- icant predictors of clearance and (b) the magnitude of the relationship with clearance. With regard to the statistically significant predictors of clearance, both proxy mea- sures indicate murders involving White victims are more likely to be cleared than those involving non-White victims. In comparison, the model using the actual clear- ance measure finds no significant difference for White as compared to non-White vic- tims. Although the actual clearance model finds a significant difference between con- tact weapons and firearms, neither of the models using proxy measures does. Three characteristics—use of a knife, home location, and more than one victim—are signifi- cant predictors in all of the models. With regard to these three characteristics, the mag- nitude of the relationship with clearance varies. In the actual clearance model, these three variables are stronger predictors of clearance than in either of the two proxy models. For example, the use of a knife as compared to a gun increases the odds of clearance by 3.5 in the actual clearance model. In the proxy models, this relationship increases the odds by 2.3 or 2.0.

Discussion and Conclusions

NIBRS data are an important source of clearance information. Unlike data used in prior studies, NIBRS clearance data provide both incident-level details and broad geo- graphical coverage. The primary goal of this research was to provide an initial explo- ration of the NIBRS clearance data to demonstrate the capabilities of these data to in- form the study of murder clearance and assess the validity of proxy clearance measures. The following discussion focuses on three issues: the factors that predict clearance using actual clearance measures from NIBRS, the validity of proxy clear- ance measures, and limitations in using NIBRS clearance data.

Analysis of NIBRS data suggests that factors related to evidence play an important role in predicting clearance, considering that knives and contact weapons as well as home location are associated with clearance. These findings are consistent with prior studies by Puckett and Lundman (2003) and Wellford and Cronin (1999), both of which used actual clearance measures collected from police records. Victim demo- graphics, though, did not appear to play a major role in explaining clearance in the actual clearance model using NIBRS data. Prior studies using proxy clearance mea- sures had found a relationship between clearance and the victim’s age and sex. In the NIBRS actual clearance model, only victim age is a significant predictor of clearance; no relationship between sex and clearance is observed. One reason for this difference may be due to using an actual clearance measure, especially given the fact that the unknown offender demographics proxy model found a relationship between victim sex and clearance.

This study also explored the validity of the clearance proxies previously used with SHR data to further an understanding of these measures. To the extent a proxy would have produced results similar to an actual clearance measure, researchers could use a proxy measure with greater confidence and capitalize on the long time-series of SHR data for analyzing clearance issues. Such a similarity is not found. An initial compari-

son indicated that the proxy measures overestimate the percentage of clearances (88% and 81%) as compared to the actual clearance measure (66%). Although this discrep- ancy alone is not conclusive of a problem, further investigation found that the cases do not have similar characteristics and patterns. This study found only three predictors (knife used, home location, and more than one victim) to be fairly robust and statisti- cally significant no matter which of the three clearance measures were used. Across the models, these three measures did have different relationships with clearance in terms of magnitude. As a result, a researcher could draw different inferences depend- ing on the clearance measure used. Because these proxies did not produce similar results to those of the actual clearance measure in terms of statistical significance or magnitude of the relationship, this study suggests exercising caution in using proxy clearance measures.

Finally, although NIBRS provides a direct measure of clearance, it has its own limi- tations in that NIBRS only collects characteristics about the crime incident. Victim and incident characteristics constitute only one aspect of clearance (for a discussion, see Wellford & Cronin, 1999). For those interested in developing police policies and practices regarding crime clearance, these factors have the least direct application because police do not have control over factors such as victim demographics, crime location, or weapon used. Future clearance research should examine ways to link NIBRS data with law enforcement agency information, such as the Law Enforcement Management and Administrative Statistics (LEMAS) survey. LEMAS is a sample sur- vey of local police departments and collects data on police personnel, expenditures, and special equipment that may be related to clearance rates (BJS, 2003). The present study attempted to link NIBRS clearance data with the LEMAS information. Because of the small agency bias in NIBRS and the oversampling of large agencies in LEMAS, very few agencies matched. As larger agencies begin to report in NIBRS format, LEMAS information could provide a valuable complement to NIBRS data.

As more agencies and states convert to NIBRS, it is important to gain a better understanding of these incident-level crime data and to use these data to inform crimi- nological research. This study provides an initial assessment of the capabilities of NIBRS data for examining murder clearance. It demonstrates the utility of these data for that purpose but also indicates additional avenues for future study. In particular, researchers should capitalize on the incident-level information for nonlethal violence and the capability for comparing clearance patterns and predictors across crimes.

Addington / Murder Data and Clearance Predictors 149

Appendix Clearance Status of Murders by

Number of Crime Victims, 2001 NIBRS Number of Crime Victims

One Two or More Total

Clearance Status N % N % N %

Arrest 872 59 325 68 1,197 61 Exceptional clearance 66 4 24 5 90 5 No clearance 544 37 127 27 671 34 Total 1,482 476 1,958

Notes

1. The term clearance is used rather than the more specific term arrest. This terminology parallels the Federal Bureau of Investigation’s practice of considering a crime cleared or solved for crime reporting pur- poses if there has been either an arrest or activity constituting clearance by exceptional means (Federal Bureau of Investigation [FBI], 2004a). Clearance by exceptional means refers to situations where a suspect has been identified, but circumstances beyond the agency’s control, such as the death of the suspect, prevent an arrest.

2. Readers interested in additional information about police response and clearance are directed to Wellford and Cronin (1999) and Puckett and Lundman (2003).

3. In addition to accuracy, these data collections also allow researchers to collect information not included in either the Supplementary Homicide Report (SHR) or National Incident-Based Reporting System (NIBRS) such as number of detectives assigned to the case or witnesses to the crime (e.g., Wellford & Cronin, 1999).

4. As discussed in more detail in the Data and Method section, readers should note that although NIBRS data provide actual clearance information for more jurisdictions than any previous clearance studies, these data do not cover the entire U.S. population.

5. Only murders that occurred in 2001 are included in this count and used in the analyses reported in this study. The NIBRS public-use files do contain crimes that were not committed during that given year because the data files include the full record on any criminal incident for which any activity was reported during that year. For example, the full record for a murder committed in 2000 with an arrest in 2001 will be included in both the 2000 and 2001 NIBRS public-use files.

6. Readers interested in more detailed information about NIBRS are directed to sources such as the Bureau of Justice Statistics (BJS; 2000) and FBI (1992).

7. In addition to collecting initial information about clearance, NIBRS data can be updated for a period of 3 years after the offense. Unlike the summary system and the SHR, NIBRS data are submitted in computer- ized form with unique incident numbers to allow incident-level updating (Addington, 2004).

8. Three agencies from Kentucky and one from the District of Columbia also submitted 2001 NIBRS data. In the District of Columbia, only the Metro Transit Police report NIBRS data (BJS, 2004). One reason for this situation is the fact that a few states do not have a state-level Uniform Crime Reporting (UCR) pro- gram (Maltz, 1999). In these cases and under special circumstances, the FBI certifies individual agencies (BJS, 1997).

9. For example, in NIBRS-certified states such as Massachusetts and Michigan, departments serving larger communities such as Boston and Detroit do not submit NIBRS data (Addington, Loftin, & McDowell, 2001). The Fairfax County Police Department (Virginia) is the largest NIBRS-reporting agency (covering a population of 948,050) (Justice Research and Statistics Association, 2005).

150 Homicide Studies

10. Although the murders must have been committed in 2001 to be included in this study, the clearance numbers include clearance activity that occurred in either 2001 or 2002. Approximately 6% of the arrests and 14% of the exceptional clearances occurred in 2002.

11. NIBRS may undercount the number of murders involving an unknown offender. The code used to indicate unknown offenders is 00, and it is possible this code is overwritten if an arrest (or exceptional clear- ance) is subsequently made. In such a situation, the agency could complete the offender details using the arrestee information (J. Jarvis, personal communication, November, 2004). These changes are not flagged so it is impossible to know the number of unknown (00) offender murders that end in arrest. It is likely this number is very small because cases in which nothing is known about the offender (not even the number of offenders involved) would be the least likely to be cleared.

12. Prior studies indicated that victim-offender relationship is a relevant clearance predictor. In this study, victim-offender relationship is excluded from the models because of the large amount of missing data. Two alternative estimation strategies were considered. One involved dropping those cases with missing victim-offender data. This strategy would have resulted in losing almost 700 cases (or 36% of the data) from the analysis and was deemed to be undesirable. A second strategy included the missing cases as an unknown category. Using unknown victim-offender relationship as a comparison resulted in nearly perfect prediction. One reason for this high correlation may be due to the practice of some agencies to overwrite the initial inves- tigation data and specify a victim-offender relationship once an arrest is made (J. Jarvis, personal communi- cation, October, 2004).

REFERENCES

Addington, L. A. (2004). The effect of NIBRS reporting on item missing data in murder cases. Homicide Studies, 8, 1-21.

Addington, L. A. (2005). The characteristics of cleared murders reported to NIBRS. In V. P. Bunge & C. R. Block (Eds.), Linking data to practice in homicide/violence prevention: Proceedings of the 2004 meet- ing of the homicide research working group (pp. 160-165). Ann Arbor, MI: Inter-university Consortium of Political and Social Research.

Addington, L. A., Loftin, C., & McDowall, D. (2001, November). The quality of NIBRS murder data. Paper presented at the annual meeting of the American Society of Criminology, Atlanta, GA.

Allison, P. D. (2002). Missing data. Thousand Oaks, CA: Sage. Bureau of Justice Statistics. (1997). Implementing the national incident-based reporting system: A project

status report. Retrieved December 18, 2002, from http://www.ojp.usdoj.gov/bjs/pub/pdf/inibrs.pdf Bureau of Justice Statistics. (2000). Effects of NIBRS on crime statistics. Retrieved December 18, 2002,

from http://www.ojp.usdoj.gov/bjs/pub/pdf/enibrses.pdf Bureau of Justice Statistics. (2003). Law Enforcement Management and Administrative Statistics (LEMAS):

2000 sample survey of law enforcement agencies (Codebook). Ann Arbor, MI: Inter-university Consor- tium for Political and Social Research. Retrieved on May 4, 2004, from http://www.icpsr.umish.edu/ NACJD

Bureau of Justice Statistics. (2004). Level of participation by states as of December, 2002. Retrieved April 30, 2004, from http://www.ojp.usdoj.gov/bjs/nibrsstatus.htm

Chilton, R., & Jarvis, J. (1999). Using the National Incident-Based Reporting System (NIBRS) to test esti- mates of arrestee and offender characteristics. Journal of Quantitative Criminology, 15, 207-224.

Federal Bureau of Investigation. (1992). Uniform crime reporting handbook, National Incident-Based Reporting System edition. Washington, DC: Government Printing Office.

Federal Bureau of Investigation. (2004a). Crime in the United States 2003. Washington, DC: Government Printing Office.

Federal Bureau of Investigation. (2004b). National Incident-Based Reporting System, 2001 [Data file]. Complied by the U.S. Department of Justice, Federal Bureau of Investigation. Ann Arbor, MI: Inter- university Consortium for Political and Social Research. Retrieved on May 4, 2004, from http://www.fbi .gov/ucr/nibrs/manuals/v1all.pdf

Addington / Murder Data and Clearance Predictors 151

Justice Research and Statistics Association. (2005). 50 Largest NIBRS Agencies. IBR Resource Center. Retrieved March 31, 2005, from http://www.jrsa.org/ibrrc/index.html

Litwin, K. J. (2004). A multilevel multivariate analysis of factors affecting homicide clearances. Journal of Research in Crime and Delinquency, 41, 327-351.

Maltz, M. D. (1999). Bridging gaps in police crime data: A discussion paper from the BJS Fellows Program. Washington, DC: Bureau of Justice Statistics.

Poggio, E. C., Kennedy, S. D., Chaiken, J. M., & Carlson, K. E. (1985). Blueprint for the future of the uni- form crime reporting program: Final report of the UCR study. Washington, DC: U.S. Department of Justice.

Puckett, J. L., & Lundman, R. J. (2003). Factors affecting homicide clearances: Multivariate analysis of a more complete conceptual framework. Journal of Research in Crime and Delinquency, 40, 171-193.

Regoeczi, W. C., Kennedy, L. W., & Silverman, R. A. (2000). Uncleared homicides: A Canada/United States comparison. Homicide Studies, 4, 135-161.

Riedel, M. (1998). Counting stranger homicides: A case study in statistical prestidigitation. Homicide Stud- ies, 2, 206-219.

Riedel, M. (1999). Sources of homicide data. In M. D. Smith & M. A. Zahn (Eds.), Homicide: A sourcebook of social research (pp. 75-95). Thousand Oaks, CA: Sage.

Riedel, M., & Jarvis, J. (1999). The decline of arrest clearances for criminal homicide: Causes, correlates, and third parties. Criminal Justice Policy Review, 9, 279-306.

Riedel, M., & Rinehart, T. A. (1996). Murder clearances and missing data. Journal of Crime and Justice, 1, 83-102.

Wellford, C., & Cronin, J. (1999). An analysis of variables affecting the clearance of homicides: A multistate study. Washington, DC: Justice Research and Statistics Association.

Lynn A. Addington is an assistant professor of justice, law, and society at American University. She is inter- ested in the nature of violent crime and its impact on victims as well as use of national measures of crime including National Incident-Based Reporting System and the National Crime Victimization Survey. Her research has appeared in outlets such as the Journal of Quantitative Criminology and Homicide Studies and has been supported by grants from the National Institute of Justice, the Bureau of Justice Statistics, and the American Statistical Association.

152 Homicide Studies