Homicide as a community problem in the United States.

The Importance of Homicide as a Community Problem in the United States

Within the realm of criminal justice, probably no offense is as studied as that of homicide. In the United States, most certainly no other offense is treated more seriously in the criminal justice system. The most severe punishments are reserved for those individuals found guilt3 of homicide, and no statute of limitations exists for homicide as is found in most other types of offenses. Homicide or fear of homicide, garners an immediate reaction from communities. As such law enforcement continually focuses efforts on the issues surrounding violent crime such as illegal gun or drug crime crackdowns. Ultimately, they hope to lessen the likelihood of homicide victimization and, as a result, fear within a community.

Patterns of Homicide

To understand the dynamics of any criminal offense, one would focus upon general patterns, and homicide is no exception. These patterns can be expressed in terms of temporal patterns (how homicide has changed over time), spatial patterns (how homicide changes over regions or locations), patterns in the inherent characteristics in terms of victims, offenders or other qualities of the homicide incident, or any combination of these qualities. To provide a thorough review of the studies conducted about homicide is beyond the scope and purpose of this paper. Suffice it to say, there have been numerous studies focusing on any one of the major themes listed above. However, in terms of recognizing patterns in the homicide data, these studies traditionally limit themselves to analyzing only a few dimensions at one time. For example, age-specific rates for victims or offenders may be analyzed over time or space. Even though theoretically possible, difficulties in interpretation arise when traditional pattern analysis techniques, such as cross-tabulations or scatter plots, are used to analyze more than three dimensions at one time. To try to view large complex data sets such as can be found with homicide in order to recognize patterns requires a more sophisticated approach. Newer technologies have made these sophisticated pattern analysis techniques more accessible through both the availability of computing resources and the ease of use.

Pattern Recognition and Data Mining

The underlying goal of pattern recognition is, ultimately, data reduction. Instinctively, humans reduce the amount of information in the world by organizing its constituents into a series of conceptual types. This is accomplished through highlighting important characteristics that define the differences and disregarding the details that add little value. The details of criminal incidents are invaluable to law enforcement to achieve the goals of solving and reducing crime. However, making sense of those details is a challenge when the scope is broader than a single incident. In order to truly use the data, the analyst must first organize the data. This can be done in one of two ways: either conceptual classification or numerical classification.

Conceptual classification methods are often used to identify ideal or polar (extreme) types and are more likely to be drawn from a collective set of experiences that represent the concepts rather than an actual set of cases. Numerical classification uses quantitative techniques to identify like cases that translate into classes. Numerical classification has been more readily used in biology and other sciences rather than in the social realm. However, that practice is beginning to change as the technology involved has become more accessible to a wider audience (Bailey 1994).

The amount of information available to law enforcement via their incident reports and external sources, such as medical examiners offices, drug laboratories, and other sources of intelligence, have placed new demands on already overstretched resources which have limited the time and attention that each law enforcement employee can spend on investigation and analysis. Law enforcement would benefit from the application of newer computer technologies in both hardware and software to help cut a clearer path through their data in order to help define problems in their communities. For these and many other reasons, the pattern recognition capabilities in data mining have become increasingly popular with law enforcement.

Crime analysts also can use pattern recognition techniques common to data mining to examine large criminal justice data sets. In the case of homicide, the Uniform Crime Reporting (UCR) Program through its Supplementary Homicide Report (SHR) collects information on the incident for the vast majority of reported homicides from state and local law enforcement. By using pattern recognition techniques, such as cluster analysis, with homicide incident information, the patterns that occur naturally within the data set can be used to provide a deeper understanding of the dynamics of homicide in the United States. This can aid more in-depth studies that focus on the underlying causes of or concurrent factors that contribute to homicide.

Objectives of the Study

By using incident-specific information available through the SHR, pattern recognition or data mining techniques can be applied to discern any patterns that exist in the homicide data. One methodology employed is cluster analysis. Cluster analysis uses a series of mathematical computations to identify groups that occur in data sets. These groups, or clusters, could be used to delve beneath the surface of the homicide rate so often quoted to reveal the nature of those rates and to explore them in a regional and temporal context. The objectives could be summarized by the following questions:

* What do the characteristics of homicide incidents reveal about the nature of homicide in the United States?

* How might the homicide data be decomposed to research it over time and space?

* How have the characteristics of homicide changed regionally since 1980? Methodology

Data

The UCR Program defines murder and nonnegligent homicide as "the willful (nonnegligent) killing of one human being by another" (USDOJ 1984, p. 6). This definition does not include suicides, accidental deaths, assaults to murder, traffic fatalities, or attempted murders. Although justifiable homicides by law enforcement officers in the line of duty or private citizens during the commission of a felony are considered willful killings, they have not been used in the calculation of published murder rates. Since 1962, (1) the FBI has collected information on homicide incidents that can be employed by a variety of users to explore the nature of homicides. This information, the Supplementary Homicide Report, is collected in addition to the official reports of crime used to calculate the national murder rate. The attributes collected on the form include the age, sex, and race of both victim and offender, weapons, relationship of victim to offender, circumstances of the homicide, as well as information on multiple victims and offenders. Additionally, law enforcement submits information on those incidents classified as justifiable homicide by UCR definitions via the SHR. It is included in this analysis.

The SHR data from 1980 through 2002 were recoded to express each incident in terms of 36 characteristics. These characteristics include the presence or absence of such things as juvenile offenders or victims, male victims or offenders, a firearm, or familial relationships between victims and offenders to name a few. More detail can be found on these characteristics in the Appendix of this study. These variables become the basis for the cluster analysis. For the remainder of the analysis, the homicide data are assigned to a county based upon the location and jurisdiction of the reporting agency. These county-level files form the basis of the geographic analysis. Because the incident-level homicide data can have fluctuations based upon the reporting history of the agencies involved as well as rare or extreme events, the data used for this study are the 3-year centered moving averages for each county. For example, the 3-year centered moving average representing levels of homicide for a county for 1982 is the average of the reports for 1981, 1982, and 1983. By using this data smoothing technique, however, the first and last years in the series are lost. The final time series used in this study represent the years 1981 to 2001.

Cluster analysis of homicides incidents to determine types

The methodology of this study focuses on two primary areas: the use of cluster analysis to detect patterns inherent in the data set itself and the mapping of those homicides to a location with the help of geographic information system (GIS) technology.

As stated previously, a cluster analysis was performed on the recoded SHR data in order to discern any pat terns inherent in the data. Cluster analysis allows for the data to drive the determination of types or groups rather than preconceived ideas of how homicides occur in the nation. It uses measurements of similarity based upon the characteristics of each homicide incident to allow "clustering" of types to be identified. Once the valid and reliable "types" of homicide are determined for each year, each reporting agency's types of homicide will be analyzed within each region to track their spatial movement and to see if they are related to one another or could be considered to be related to population movements. For a more thorough discussion of the technique, including the checks on the validity and reliability of the results, a Technical Note is available upon request from the FBI's Crime Analysis, Research and Development Unit, telephone [phone].

GIS analysis of homicide incidents

In addition to cluster analysis, the homicide information was tracked through space with the aid of GIS technology. Initially, the location of the incident-level homicides is determined based upon the county in which the reporting agency resides. These incidents are spatially attributed to the point of the county centroid, which is the spatial center of the county. The mean center weighted by the number of reported homicides for each year, as well as by cluster type, is calculated for the Nation and each of the four regions. The mean center is the geographic equivalent to the average in a data set, and can be thought of as the "balancing point." Additionally, the standard deviation for the mean center is calculated for each year to indicate the dispersion of the data. This standard deviation is calculated for both the x- and y-dimension in space and is represented by an ellipse. For a more thorough discussion of the calculation of the weighted mean center and standard deviational ellipse, a Technical Note is available upon request from the FBI's Crime Analysis, Research and Development Unit, telephone [phone].

Incidents of Homicide in the United States

Spatial trend of homicide incidents from 1981 to 2001 for the Nation

An examination of the calculated mean centers for the reports of incident-level homicide revealed that they appeared to be balanced near the geographic center of the United States. However, the standard deviational ellipse for the 2001 centered value indicated that the majority of the homicide incidents drifted to the east coast, and the reporting population was more geographically dispersed than the reports of homicide. This is primarily reflected in the northwest extant of the standard deviational ellipse. In other words, homicides were more concentrated than the reporting populations of the same counties for the same time period (2001). Additionally, there appears to be an urban bias to the homicide incidents shown in this map. Approximately 66 percent of the Nation's homicides took place in an area that encompassed nearly 51 percent of the urban places with a population of 100,000 or more. (See Figure 5.1.)

[FIGURE 5.1 OMITTED]

The study period (1981-2001) began with a murder rate of approximately 9.8 murders per 100,000 inhabitants. However, the murder rate declined in subsequent years. The rate again rose to the same high in 1991 and then declined to the lowest point in 21 years by 2000 to 5.5 murders per 100,000 inhabitants (USDOJ 2003). Over time, the population covered by agencies reporting SHR data has shown a nearly true westerly progression from east to west. Interestingly, the Census Bureau reported the same momentum for the U.S. population. During the same time period, the incidence of homicide appears to have a western bias in its trend, and there is less movement for reported homicides. When the homicide rate is taken into consideration, there is a more easterly bias in higher crime rate years than the lower crime rate years in the latter part of the study period. This may be a reflection of a decline in the homicide rates in the East. (See Figure 5.1, Inset.)

Spatial trend of homicide incidents from 1981 to 2001 for each region

The UCR Program divides the United States into four geographic regions for data analyses: the Northeast, the Midwest, the South and the West. When the homicide data are analyzed by region, there appears to be strong evidence that the level of urbanity is tied to the incidence of homicide on a regional level, as well. Visually, the data show that urban areas are in a more geographically dispersed pattern in the Midwest and South. Although the West accounts for approximately 40 percent of the Nation's populated places with 100,000 or more in population, they are almost all concentrated around San Francisco and Los Angeles. In the Northeast, these larger urban centers are almost all concentrated around the New York City and Boston areas. Additionally, the Northeast proportionally contributes only about 10 percent to the total number of the larger urban centers for the Nation. (See Figure 5.2.)

[FIGURE 5.2 OMITTED]

The regional dispersion of homicides in relation to reporting population nearly mimics the results of the national analysis. Except for the Northeast, homicides tend to be more concentrated than reporting population over time. With the exception of the South, most mean centers are also geographically close to major urban centers of the region (for example, New York City, Chicago/Detroit, and Los Angeles/San Francisco). The South shows a much wider dispersion of homicide, which may reflect the geographic pull of more widely dispersed urban centers (for example, New Orleans, Washington, D.C., Houston, and Atlanta). (See Figure 5.2.)

This urban pull is also reflected in the geographic progression of the mean centers through time. Although the reporting population of the Midwest appears to split the difference between Chicago and Detroit, there appears to be a bias towards Chicago in terms of the homicide reports. The reporting population in the Northeast appears to be geographically stagnant near New York City. However, the incidence of homicide drifts westerly during the study period. The reporting population mean center of the West has been moving in a southerly direction towards Los Angeles, but homicides show a southeasternly pull between Los Angeles and Las Vegas. Again, the South does not show a bias towards any particular urban area with the reporting population moving towards the northwest and homicide incidents moving towards the northeast. In general, one does not see evidence that the incidence of homicide appears to be linearly related to the reporting population during the study period.

Characteristics of Detected Homicide Clusters

The results of the cluster analysis produced two highly robust types of homicide that can be tracked for the entire study period. The first robust cluster is comprised primarily of incidents that involve an unknown offender, and as such, much of the information about the offender and the circumstances surrounding the incident are reported as unknown. The remaining incidents are either clustered together as one all other type, or in some years, the remaining cases are divided based primarily on the race of the offender. In those years, there are two remaining clusters: one with black offenders and one with white and all other race (2) offenders. Given the consistency (reliability) of the results from each year, the clusters produced by this analysis will form the basis of the remaining analyses.

The category unknowns accounts for approximately 30 percent of the 417,505 homicides reported through the SHR during the study period. These are the homicides in which little to nothing is known about the offender at the time of the incident report. There is a slight urban bias to these unknowns, and the victims are more likely to be black, male, and adult than the rest of the victims of the homicides. However, when the unknowns are compared to all other homicides, both groups appear to be very similar in terms of weapons used. (See Table 5.17.) Over the length of the study period, there appears to be a slight increase in the proportion of unknowns. (See Figure 5.3.)

[FIGURE 5.3 OMITTED]

Regionally, the data reflected many of the same patterns concerning age and weapons associated with the two identified types of homicide. However, some striking differences are obscured by the national figures. In the West, one sees a significant increase in the proportion of white victims (65 and 70 percent, respectively for unknowns and all other homicides) when compared to the national figures (46 and 51 percent, respectively). Additionally, both the South and the West show a higher incidence of each type of homicide in suburban areas than do the remaining regions. In general, although many of the differences between the two types are subtle, the South also showed much more similarity between the unknowns and all other homicides than did the other regions. (See Table 5.17.) Since 1981, the proportion of unknowns on a regional level does not consistently track with fluctuations in the murder rate. In the Midwest, the South, and the West, the percent of unknowns either grew or remained stable during times of declining murder rate. It was only in the Northeast that the trends of the murder rate and percentage of unknowns appear to move in congruent directions (See Figure 5.4.)

[FIGURE 5.4 OMITTED]

Spatial Occurrence of Homicide Clusters

Spatial trend of homicide clusters from 1981 to 2001 for the Nation

Unknowns show a lot more geographic progression in a westerly direction during the study period than do all other homicides. In the early part of the study period, the mean center for unknowns appeared more to the east than the mean center for the remaining homicides. That bias is almost nonexistent by the end of the study period. For the entire study period, the unknown offender homicides also show a more northern bias than other homicides. The standard deviational ellipse for the centered average value for 2001 (the most current data year in the study) shows a wider dispersion for the unknowns on the east-to-west axis. However, the remaining homicides are more dispersed on the north-to-south axis. This indicates a more concentrated band of unknowns that incorporates the effects of urban areas in the northern part of the Northeast (Boston/Connecticut) as well as in southern California. The all others seem to incorporate the effect of more southerly urban areas such as Atlanta, New Orleans, and Houston. (See Figure 5.5.)

[FIGURE 5.5 OMITTED]

Spatial trend of homicide clusters from 1981 to 2001 for each region

Within the Nation's four regions, the data reflected a similar pattern where each type of homicide has distinct mean centers. These differences for the two types reflect the influence of diverse communities even within the same region. All regions but the West showed the same spatial progression through time for the two types of homicide. In all cases, this progression runs counter to the progression for the reporting population. Interestingly, unknowns show an almost true southerly progression in the West while all other homicides drift to the northeast. The standard deviational ellipse for the most recent data year (the centered average value for 2001) showed that the dispersion of unknowns was less than or equal to the remaining homicides in all regions but the South. (See Figure 5.6.)

[FIGURE 5.6 OMITTED]

Discussion and Conclusion

The differing rates of homicide amongst the regions, particularly in the South and West, have been noted in the past by law enforcement and researchers alike. However, instead of these differences in the levels of homicide being a result of global processes within a region, there appears to be evidence of more subtle processes that are connected to local urban centers. The regional incidence of homicide may be a reflection of the level of urbanity or change in urbanity rather than strictly the numbers of people that reside there. Since the patterns of growth and decline vary by region, this could be one explanation for the varying results in movement and types of homicide among regions.

The unknowns are difficult to draw too many conclusions about, since, by definition, little is known of the circumstances surrounding the homicide. However, it could easily be seen how these homicides may differ qualitatively from the remaining homicides that are often between people who know one another and in many cases are the result of arguments. The regional differences in the trends and proportions of these unknown homicides appear to be influenced by communities different from the remaining homicides. Again, this points to the dominance of communities different from those that drive the trends for the remaining homicides. The results of this analysis show that there is a definite need for further exploration of what is driving regional trends in homicide. The evidence seems to point to a complex interaction between regional differences in the underlying factors affecting homicide and the regional differences in the types of homicide itself.

Appendix

Variable                     Description

MSA                          MSA location
Suburban                     Suburban location
Single Victim                Incident involved only a single victim
Single Offender              Incident involved only a single offender
Unknown Offender             Incident involved an unknown offender
Juvenile Victim              Victim under the age of 18 years old
Male Victim                  Male victim involved
Female Victim                Female victim involved
Unknown Victim               Unknown victim involved
White Victim                 White victim involved
Black Victim                 Black victim involved
AIAN Victim                  American Indian or Alaskan Native victim
                               involved
AOPI Victim                  Asian or Pacific Islander victim involved
Juvenile Offender            Offender under the age of 18 years old
                               involved
Male Offender                Male offender involved
Female Offender              Female offender involved
Unknown Offender             Unknown offender involved
White Offender               White offender involved
Black Offender               Black offender involved
AIAN Offender                American Indian or Alaskan Native offender
                               involved
AOPI Offender                Asian or Pacific Islander offender
                               involved
Firearm used                 Includes handgun, rifle, shotgun, other
                               gun, and general firearm
Other Serious Weapon         Includes knife/cutting instrument, blunt
                               object, personal weapons, poison,
                               pushed/thrown out of window, explosives,
                               fire, narcotics/drugs, drowning,
                               strangulation, asphyxiation, and other
Intimate Relationship        Includes husband, wife, common-law
                               husband, common-law wife, boyfriend,
                               girlfriend, ex-husband, ex-wife, and
                               homosexual relationship
Other Family Relationship    Includes mother, father, son, daughter,
                               brother, sister, in-law, stepfather,
                               stepmother, stepson, stepdaughter, and
                               other family
Otherwise Known              Includes neighbor, acquaintance, employee,
                               employer, friend, and otherwise known to
                               victim
Not Known to Victim          Includes stranger.
Unknown Relationship         All instances where relationship of victim
                               to offender cannot be determined.
Felony Type                  Includes rape and robbery
  Circumstance - Violent
Felony Type                  Includes narcotic drug laws
  Circumstance - Drug
Felony Type                  Includes burglary, larceny, motor vehicle
  Circumstance - Other         theft, arson, prostitution and
                               commercialized vice, other sex offenses,
                               abortion, gambling, and other-not
                               specified
Other Circumstance -         Includes lover's triangle, brawl due to
  Arguments                    influence of alcohol, brawl due to
                               influence of narcotics, argument over
                               money or property, and other arguments.
Other Circumstance -         Includes gangland killings, juvenile gang
  Organized                    killings, and institutional killings.
Other Circumstance - Other   Includes child killed by babysitter,
                               sniper attack, and other.
Suspected Felony             Circumstances indicate possible felony
                               type murder, but sufficient facts to
                               identify type of felony not available.
Justifiable Homicide         The intentional killing of a person
                               without evil design and under such
                               circumstance of necessity or duty as to
                               render the act proper. Includes felons
                               killed by either private citizen or
                               police.

The preceding list reflects particular characteristics captured in a
homicide incident reporting through the SHR. These characteristics
were recoded to reflect whether or not that characteristic was present
or not present. If the characteristic was present on the incident,
that variable was coded as the value 1. Otherwise, the variable was set
to the value of 0. The cluster analysis algorithm described in detail
in the Technical Note used these values in its calculations. The
Technical Note is available upon request from the Crime Analysis,
Research and Development Unit, telephone [phone].

References

Aldenderfer, Mark S. and Roger K. Blashfield. (1984). Cluster Analysis. Newbury Park, CA: Sage Publications.

Bailey, Kenneth D. (1994). Typologies and Taxonomies: An Introduction to Classification Techniques. Newbury Park, CA: Sage Publications.

Goldsmith, Victor, Philip G. McGuire, John H. Mollenkopf, Timothy A. Ross, eds. (2000). Analyzing Crime Patterns: Frontiers of Practice. Thousand Oaks, CA: Sage Publications, Inc.

Langworthy, Robert H. and Eric S. Jefferis. (2000). The Utility of Standard Deviation Ellipses for Evaluating Hot Spots. In Goldsmith, Victor, Philip G. McGuire, John H. Mollenkopf, Timothy A. Ross, (Eds.), Analyzing Crime Patterns: Frontiers of Practice. Thousand Oaks, CA: Sage Publications, Inc.

Levine, Ned. (May 2002). CrimeStatII: A Spatial Statistics Program for the analysis of Crime Incident Locations. Houston, TX: Ned Levine & Associations and Washington, D.C.: the National Institute of Justice.

McCue, Colleen, Emily S. Stone, and Teresa P. Gooch. (2003). "Data Mining and Value-Added Analysis, Law enforcement Bulletin. November, p.1-5.

Romesburg, H. Charles. (1984). Cluster Analysis for Researchers. Belmont, CA: Lifetime Learning Publications.

SPSS. "The SPSS TwoStep Cluster Component," Technical Report downloaded from www.spss.com on April 1, 2004.

U.S. Department of Justice. Federal Bureau of Investigation (2003). Crime in the United States, 2002. Washington, D.C.

U.S. Department of Justice. Federal Bureau of Investigation (1984). Uniform Crime Reporting Handbook. Washington, D.C.

Endnotes

(1) Supplementary homicide information has been collected since the beginning of the UCR Program in the early 1930s. However, this information was not made available for general dissemination until 1962 and has gone through various revisions since that time. The data used for this analysis reflect the latest version of information collected which has remained the same since 1980.

(2) These race categories include Asian and Other Pacific Islander and American Indian and Alaskan Native.

Table 5.17
Characteristics of detected homicide clusters

Percent of total

                                  Nation                Northeast

                            Unknowns   All Other   Unknowns    All Other

MSA status                    94.2       84.8        98.3        94.7
Suburban                      18.1       21.7         8.3        17.1

Black Victim                  50.6       45.9        53.1        49.2
White Victim                  45.9       51.4        43.3        48.2

Female Victim                 19.8       24.3        17.1        24.8
Male Victim                   81.3       77.4        83.7        76.9

Juvenile Victim               7.0        10.7         6.7        11.8

Firearm Used                  65.4       64.5        66.6        55.7
Other Serious Weapon Used     34.6       35.6        33.4        44.3

                                 Midwest                 South

                            Unknowns   All Other   Unknowns   All Other

MSA status                    96.1       90.4        89.7       76.3
Suburban                      11.1       15.0        22.8       24.8

Black Victim                  67.8       57.8        56.2       51.5
White Victim                  31.1       40.7        41.6       47.1

Female Victim                 21.2       25.3        21.4       24.0
Male Victim                   80.5       76.5        79.4       77.5

Juvenile Victim                7.4       12.6         5.8        8.7

Firearm Used                  67.1       63.2        64.2       68.4
Other Serious Weapon Used     32.9       36.9        35.8       31.8

                                   West

                            Unknowns   All Other

MSA status                    95.5       90.3
Suburban                      25.8       24.0

Black Victim                  27.7       23.4
White Victim                  65.3       70.3

Female Victim                 18.8       23.8
Male Victim                   82.4       78.0

Juvenile Victim                8.6       12.0

Firearm Used                  64.7       63.7
Other Serious Weapon Used     35.3       36.3