Bridging Gaps in Police Crime Data A Discussion Paper from the BJS Fellows Program by Michael D. Maltz Department of Criminal Justice University of Illinois at Chicago and Visiting Fellow Bureau of Justice Statistics September 1999, NCJ 176365 This paper is based on a Workshop on Uniform Crime Reporting Imputation, sponsored by the Bureau of Justice Statistics and the Federal Bureau of Investigation Uniform Crime Reporting Program U.S. Department of Justice Bureau of Justice Statistics Jan M. Chaiken, Ph.D. Director The points of view or opinions expressed in this document are those of the author and do not necessarily represent the official position or policies of the U.S. Department of Justice. BJS Discussion Papers promote the exchange of information, analysis, and ideas on issues related to justice statistics and to the operations of the justice system. This document was prepared under cooperative agreement 95-BJ-CX- 0001 for the BJS Visiting Fellowship Program. The author may be reached at the following addresses: Professor Michael D. Maltz Department of Criminal Justice University of Illinois at Chicago 1007 W. Harrison Street (M/C 141) Chicago, IL 60607-7140 email: mikem@uic.edu Acknowledgments I began work on this project because, although I had been using the UCR for many years, I had never understood all of its intricacies -- and felt somewhat embarrassed to ask simple questions about why certain procedures were used, because obviously everyone else knew. It turned out, however, that most people seemed to be as much in the dark as I was, perhaps about different aspects of the UCR,and during this project we began to share our knowledge, each of us having an understanding of different aspects of the data collection and analysis process. This report is, then, more a collaboration than a single-authored effort, a kind of "open source" presentation of our collective knowledge. [Although the knowledge is collective, the interpretation of that knowledge is my own.] The information contained herein is based on a 2-day meeting held over 2 years ago; analyses of UCR data conducted since then; and conversations, letters, faxes, and e-mails between me and a number of colleagues. In particular, I wish to acknowledge the comments and advice of the following: * From BJS -- Jan Chaiken, Larry Greenfeld, Tom Hester, Charles Kindermann, Pat Langan, Sue Lindgren, Don Manson, Marilyn Marbrook, Mona Rantala, Steve Smith, Paul White, and Marianne Zawitz * From the FBI -- Yoshio Akiyama, Bennie Brewer, Kenneth Candell, Carlos Davis, Gil Gee, Antonio Hwang, Dawn Kording,Vicki Major, Jim Nolan, Sharon Propheter, and Maryvictoria Pyne * From the research community -- Dan Bibel, Becky Block, Roland Chilton, Chris Dunn, Bob Flewelling, Jamie Fox, John Jarvis, Jim Lynch, Mike Maxfield, and Howard Snyder. In addition, I wish to thank the UCR program personnel from the 50 States with whom my research assistants (Leanne Brecklin, University of Illinois at Chicago; Chris Kenaszchuk, University of Maryland; and Todd Minton and Matt Durose, BJS) and I corresponded and spoke. I received cooperation from officials in every State in putting together this report; I hope that the end result is of use to them and to others who deal with crime statistics. This report was written while I was a Visiting Fellow at the Bureau of Justice Statistics and completed while I was on sabbatical from the University of Illinois at Chicago. While I greatly appreciate the help I received from BJS, FBI, and State officials, they should not be held responsible for any errors in this report, and the opinions, conclusions, and recommendations expressed herein are my own and should not be construed as the policy of any of these organizations. Michael D. Maltz Department of Criminal Justice University of Illinois at Chicago Visiting Fellow Bureau of Justice Statistics Summary Crime in the United States (CIUS), published annually by the FBI, is a compilation of the Uniform Crime Reports (UCR) provided by over 18,000 policing jurisdictions. It represents one of the two primary sources of data about crime in the United States, the National Crime Victimization Survey (NCVS) being the other. While the NCVS is a very reliable indicator of national trends in crime, it is based on a survey of under 50,000 house- holds and thus cannot provide local information on crime, which is provided by the UCR and CIUS. [For a thorough understanding of the differences between the two statistical series, see Biderman and Lynch's (1991) Understanding Crime Incidence Statistics: Why the UCR Diverges from the NCS. The NCS, or National Crime Survey, was the predecessor to the NCVS. A briefer explanation can be found in The Nation's Two Crime Measures, found at http://www.ojp.usdoj. gov/bjs/abstract/ntmc.htm and included annually in CIUS.] Not only does CIUS provide local information about crime incidence, it also compiles arrest data from these jurisdictions; these data permit us to form a picture of who is committing crime (or at least, who is arrested for committing crime). The quality of the data provided to the FBI, however, is uneven. Reporting to the FBI remains for many jurisdictions a voluntary activity; although many States now mandate that agencies report crime and arrest data to them (which they then forward to the FBI), even in those States local agencies do not always comply. Moreover, despite the efforts of the FBI to maintain their quality, there are many gaps in the data that make their use questionable. While this has had limited impact in the past, the fact that the UCR data have, for the first time, been used to allocate Federal funds brings issues about data quality to center stage. In addition, the FBI is moving to implement an improved crime and arrest reporting system, the National Incident-Based Reporting System (NIBRS), to augment the summary UCR data published in CIUS. It is hoped that the study of deficiencies in UCR data will be of use in planning for the full implementation of NIBRS. This report describes the history of the UCR system and the data problems that it deals with in reporting crime, arrests, and homicide. It describes the procedures used by the FBI to fill in gaps in the data when they exist and makes suggestions about how they might be improved. Contents I. Introduction Why We Need to Look at the UCR The Information-Gathering Process Report Organization II. UCR History and Coverage State-Level Reporting Comparing Crime Data Coverage Gaps and Imputation The UCR and Funding Decisions The UCR and Electronic Access Use of Sub-National UCR Data The UCR and NIBRS III. Incomplete Crime Data Error Checking Reasons for Incomplete Reporting IV. Incomplete Arrest Data V. Processing and Publishing the Crime Data Publishing the UCR Archiving the UCR Data File VI. Procedures Used for Imputing Crime Data FBI Imputation Procedures for Crime NACJD Imputation Procedure Imputation Procedures for Arrests Imputation and "Zero-Population" Agencies Updating UCR Files VII. Inaccuracies Produced by the Imputation Procedures Incomplete-Reporting Agencies Non-Reporting Agencies "Zero-Population" Agencies Summary VIII. Suggested Imputation Philosophy Suggested Imputation Zero-Population Agencies Incomplete and Non-Reporting Agencies IX. Supplementary Homicide Reports Uses of the SHR Incomplete Provision of SHR Data by Police Departments Updating SHR Files Availability of SHR Data Sets SHR Imputation Weighting the Victim File Weighting the Offender File Problems with This SHR Imputation Procedure Suggested Alternative SHR Imputation Procedure X. Conclusions and Recommendations Reporting Practices Publishing and Archiving Imputation NIBRS Appendix A. Persons Attending the Workshop on UCR Imputation Procedures Appendix B. State On-Line Publication of Crime Data Appendix C. Characteristics of State UCR Collection Programs Appendix D. Extent of UCR Data Coverage, Alabama-Wyoming, 1958-97 Appendix E. Missing Data in UCR Files Used for the 1996 LLEBG Formula Calculations Background Extent of Missing Data, by State Characteristics of Agencies with Less than 36 Months of Violent Crime Data Agencies with 0 Months of Data Jurisdictions with between 1 and 35 Months of Data Impact of Incomplete Data Addendum on "Zeropop" Agencies References Glossary I. Introduction The Uniform Crime Reporting Program (UCR) of the Federal Bureau of Investigation (FBI) has been collecting crime and arrest data from police departments throughout the United States since 1930. (Footnote: See Glossary for definitions of many abbreviated names or terms.) The data are published in the annual report, Crime in the United States (CIUS), and represent one of the more widely used sources of longitudinal data in the social sciences. The UCR is based on monthly summary reports of crimes known to the police and arrests made by the police, that are provided to the FBI by over 17,000 of the more than 18,000 police agencies in the United States and its territories. (Footnote: Police agencies also report on other topics to the FBI, including hate crimes, personnel statistics, and law enforcement officers killed and assaulted. These topics are not covered in this report.) The FBI office that deals with the UCR is the Program Support Section (PSS), a section of the Criminal Justice Information Services (CJIS) Division. Five of the eight units within PSS are concerned with various aspects of CIUS: *The Statistical Unit collects, checks, and manages the data coming in from the police agencies. *The Communications Unit is involved in publications and data dissemination. *The Education and Training Services Unit trains local agencies in UCR data collection procedures. *The Crime Analysis Research and Development Unit analyzes data and develops specifications for new methods of presenting the data. *The CJIS Audit Unit performs quality assurance reviews to maintain the quality of the UCR. The UCR includes a Crime Index, a count of certain specific crimes occurring over the past year in each jurisdiction. These are called "Index crimes," and, listed in order of their presumptive seriousness, are murder and nonnegligent manslaughter, forcible rape, robbery, aggravated assault, burglary, larceny-theft, and motor vehicle theft. Arson was added to the Crime Index in 1979 although it is not as likely as the other Index crimes to be reported to the police, because arsons are often categorized as "fires of suspicious origin." Except for arson, these particular crimes were chosen because they were frequent, generally serious in nature, and most likely to be reported to the police; victims, their relatives, and/or bystanders who witness the incident are likely to know that incidents of those types are criminal in nature and are likely to report them. Although the UCR has some limitations (indeed, the aim of this report is to address some of them), even these limitations provide important information. For example, incomplete citizen reporting to the police of certain types of crimes has been used as an indicator of a number of police-related factors: how the relationship between offender and victim affects citizen reporting of crime; the extent to which citizens trust the police; and the effect of police policies and problems on reporting behavior. Yet the public is generally unaware that the UCR system is essentially a voluntary system; there is no federal legislation that requires states or local jurisdictions to report their crime data to the FBI. The voluntary nature of the UCR, of course, affects the accuracy and completeness of the data. Although the FBI devotes a great deal of attention to the quality of the data it publishes in CIUS, it cannot mandate agencies to provide data on time (or at all). As a consequence, the FBI must deal with problems of missing or late data, and has developed a mechanism to account for these gaps: it imputes (or estimates) data where gaps exist, which limits the accuracy of the estimated crime statistics published in CIUS. Why We Need to Look at the UCR Despite these problems with the data, adjustments for missing data have not been of major consequence in the past, since the primary purpose of the data was to present national and State trends -- and estimates were adequate for this purpose. Researchers, police administrators, and some journalists are aware of the limitations of the UCR, but it mattered little to others outside the field. However, in the recent past four changes were made in the environment in which the UCR data are being employed: *UCR data are being used to allocate Federal funds. *The data are now instantly accessible on the Internet. *Because of the greater accessibility of the data, researchers are increasingly analyzing UCR statistics at sub-national levels, but the results of their analyses may be suspect because of the way missing data are handled. * A new reporting system (the National Incident-Based Reporting System or NIBRS) now being implemented to augment the summary UCR data will increase the amount of data collected on each crime and arrest. Thus, the collection, analysis, and publication of crime data are now occurring in a new environment, due to changes in legislation, changes in the ease of access by citizens and researchers to the data, and changes in crime reporting. This means that the FBI's imputation procedures, which were adequate for handling many of the weaknesses in the current data collection system, may have to be revised. Toward this end, a Workshop on UCR Imputation Procedures was held in Washington, DC, April 24-25, 1997, and attended by key personnel from the FBI and the Bureau of Justice Statistics (BJS), as well as by researchers familiar with UCR data and their problems. The list of attendees is given in Appendix A. Just prior to the workshop the FBI had moved the Program Support Section to Clarksburg, West Virginia. The move resulted in a turnover of personnel and equipment. The workshop thus came at an opportune time for the FBI, which recognizes the need to update the procedures it has been using for over 40 years -- when the UCR had its last major revision (FBI, 1958). The workshop provided an opportunity for statisticians and researchers from both of these Federal agencies and from the user community to discuss ways of improving UCR data collection and estimation procedures. The goal of the workshop was to recommend new ways to ensure that the American public is provided with the best possible police- collected information related to crime and criminality, and to move toward that end in the most expeditious and feasible way possible. This report is based on the findings and discussions from that workshop. Issues relating to standard UCR data (i.e., crime counts, arrests) were not the only topics addressed at the workshop. Attention was also devoted to the Supplementary Homicide Reports (SHR), forms filled out by police departments that provide a more detailed description of each homicide than just the raw statistics of number of homicides. The workshop explored how these data could be made more useful, and this report discusses those findings as well. Issues related to Federal crime data are not included in this report. Thus, the accuracy or completeness of the statistics of crimes committed on Indian reservations, military installations, and national parks are for the most part excluded. This report also includes information gathered from State criminal justice agency personnel and data analyses subsequent to the workshop. The voluntary nature of the UCR system means that there is a high degree of State-to-State variation in UCR reporting. Specifically, some States mandate reporting and require reports to be channeled through (and checked by) State agencies before being transmitted to the FBI, while in other States individual jurisdictions report directly to the FBI. Although the FBI institutes quality control checks on the data it receives, the lack of uniform reporting standards and procedures results in a lack of uniformity in the Uniform Crime Reports. (Footnote: The lack of uniformity is due primarily to variation in completeness of State reporting, not to variation in what is reported. The PSS Education and Training Services Unit works with individual police agencies to ensure uniformity in reporting practices.) The Information-Gathering Process Some of the material included herein is based on informal conversations with FBI and BJS personnel and State officials who use or collect the data, and some of their statements about the UCR (or my interpretations of what they meant) may be in error. Although I have tried to verify all statements, some errors may have slipped through. Should a reader find mistakes in this report, please notify me (mikem@uic.edu), and corrections will be added to an errata sheet that will be posted on the BJS website. It seems that every decade or so I look into the intricacies of crime data (Maltz, [1972] 1999; 1984: 141) and find the following caution about official statistics from Josiah Stamp (1929: 258) applicable: "The individual source of the statistics may easily be the weakest link. Harold Cox tells a story of his life as a young man in India. He quoted some statistics to a Judge, an Englishman, and a very good fellow. His friend said, "Cox, when you are a bit older, you will not quote Indian statistics with that assurance. The Government are very keen on amassing statistics -- they collect them, add them, raise them to the nth power, take the cube root and prepare wonderful diagrams. But what you must never forget is that every one of these figures comes in the first place from the chowty dar [village watchman], who just puts down what he damn pleases." While strides have been made in improving the coverage and accuracy of police-reported crime data (in India as well as in this country), there is still need for a great deal of improvement. My hope is that this report helps to realize this goal. Report Organization The organization of this report is as follows: The next section gives a brief summary of how the coverage of the UCR has increased over the past few decades, both in terms of population covered and State collection efforts. Section III describes the reasons for incomplete crime data and Section IV problems with arrest data. Section V documents the steps necessary to verify and publish CIUS. The imputation procedures used by the FBI to account for these gaps are described in Section VI, and the problems with these imputation procedures in Section VII. Some suggested changes in the imputation procedures are described in Section VIII. Issues related to the SHR data are addressed in Section IX. Conclusions and recommendations are found in Section X. Five appendixes are included: Appendix A lists the attendees at the BJS/FBI workshop. Appendix B is a compendium of crime-related data available on the Internet from State agencies. Appendix C lists some of the characteristics of State UCR collection programs. The crime reporting history of each State is charted in Appendix D. Appendix E, written by Sue Lindgren of BJS, describes the procedures used to account for missing data in calculating the Local Law Enforcement Block Grant funding for each jurisdiction. II. UCR History and Coverage The International Association of Chiefs of Police (IACP) created the UCR in January 1930. It was created in large part to forestall newspapers from manufacturing "crime waves" out of thin air (IACP, 1929; Maltz, 1977). A national system of crime reporting, it was felt, would put the inevitable (and unpredictable) swings of crime incidence in a single jurisdiction into a proper context, reducing the media pressure put on any particular jurisdiction or police chief. This pressure has led to police departments "cooking the books" and reducing the amount of crime they recorded instead of the amount of crime reported to them. At the request of the IACP, the FBI assumed stewardship of the UCR in 1930, soon after it started. Police departments that provided crime (and other) data sent the data directly to the FBI, which compiled the data and published periodic reports. [A note on terminology: the FBI identifies all police and other agencies that report crime data with an ORI (for ORIginating Agency Identifier) number. In this report I use the terms "ORI," "reporting agency," "police department," and "jurisdiction" interchangeably.] Initially there was not enough coverage of the entire United States to permit estimation of the crime rate for the Nation as a whole. From 1930 through 1957 the FBI published the data in tables according to size of the reporting jurisdiction and did not aggregate the data to the national level. In 1958, based on a review of the UCR by a consultant committee (FBI, 1958), it was felt that there was enough coverage to begin to estimate annual crime rates for the Nation as a whole, which the FBI began to do with the publication of the 1958 report. State-Level Reporting In the late 1960's a few States that had been compiling their own statistics arranged with the FBI to act as the data collection point for all ORIs within the State, and began to send the entire State's data directly to the FBI, in an effort to make the process more manageable for the FBI. Other States also began to compile their crime statistics; under the Law Enforcement Assistance Administration (LEAA), funds were made available to the States to establish statewide programs as part of State criminal justice Statistical Analysis Centers (SACs). The SACs were developed in an effort to build analytic capability in States, so that they could deal with their crime problems by themselves. The FBI then developed requirements for State UCR collection programs; currently, 44 States have met these requirements and send all of their agencies' data to the FBI.( Footnote: The decrease in the early 1980's was due to the loss of LEAA funding.) See figure 1, which shows how the program developed over the past four decades. ---------------------------------------------------------------- Figure 1. Number of States Submitting UCR Data Directly to the FBI Number of States 1969 2 1970 9 1971 11 1972 13 1973 22 1974 32 1975 36 1976 41 1977 43 1978 45 1979 47 1980 44 1981 40 1982 39 1983 41 1984 41 1985 41 1986 41 1987 41 1988 41 1989 42 1990 42 1991 42 1992 42 1993 44 1994 44 1995 44 1996 44 1997 44 Source: CIUS, 1969-97, and States responding by letter, e-mail, and telephone ---------------------------------------------------------------- Although some of the SACs disappeared or were scaled down after Federal funding declined, States have continued to compile their own crime data. Most States and territories have set up State-level UCR programs --some under SACs, some under the State police, and some in other agencies -- and now publish annual crime reports, and the FBI continues to compile and publish the data for the Nation as a whole. Appendix B provides a (partial) listing of the availability of State on-line publication of crime data. Comparing Crime Data The annual publication of CIUS is still an occasion for the media to compare a jurisdiction's crime rates with those of other jurisdictions and with its past experience, so media pressure has not been entirely dispelled. But nowadays reporters and the public are more sophisticated and recognize that the police have only a limited ability to affect many types of crime. This has eased the pressure on police administrators "to keep the crime rate down" by reporting less crime than had occurred, although the pressure to falsify crime data still persists, from Philadelphia to New York City to Atlanta to Boca Raton. (Footnote: See The Philadelphia Inquirer, November 1, 1998, "How To Cut City's Crime Rate: Don't Report It;" The New York Times, August 3, 1998, "As Crime Falls, Pressure Rises To Alter Data;" The Atlanta Journal-Constitution, May 21, 1998, "Manipulation of Crime Figures Alleged;" and The Miami Herald, May 3, 1998, "Sugarcoating? Officer Faked Boca Crime Stats.") Other problems with the data are due to inadequate systems and procedures. But as local jurisdictions automate their crime record systems we should be able to see improvements in crime data accuracy. New software products for police records management include provisions for automated reporting of UCR and NIBRS data in the correct formats. And the accuracy of crime data seems to have been improving. As figure 2 shows, the UCR estimates of violent crime recorded by the police have been drawing closer to (and following the same general pattern of) estimates of violent victimizations that victims say they reported to the police (the two middle lines). ---------------------------------------------------------------- Figure 2. Four measures of serious violent crime Serious Violent Crimes (including homicide) Total Victimizations Crimes violent reported recorded by Year crime to the police the police Arrests 1973 3,590,500 1,861,000 715,900 392,700 1974 3,800,000 2,030,000 791,000 462,900 1975 3,594,700 1,976,100 844,100 441,100 1976 3,613,300 2,039,600 822,800 414,600 1977 3,662,500 1,966,800 845,700 438,500 1978 3,626,100 1,879,800 894,100 469,900 1979 3,834,800 2,020,500 998,700 467,700 1980 3,794,400 2,037,500 1,108,300 482,900 1981 4,101,700 2,217,900 1,125,000 496,600 1982 3,926,200 2,232,700 1,108,300 547,400 1983 3,455,500 1,877,400 1,064,900 516,600 1984 3,683,300 2,003,500 1,081,800 501,600 1985 3,358,400 1,926,500 1,126,700 506,800 1986 3,284,700 1,900,200 1,268,800 565,000 1987 3,424,900 1,987,800 1,265,300 568,100 1988 3,563,000 1,942,000 1,336,700 600,000 1989 3,533,700 1,847,800 1,405,000 666,100 1990 3,500,600 1,949,200 1,556,800 722,400 1991 3,712,000 2,133,000 1,632,700 738,200 1992 3,987,000 2,161,000 1,657,300 722,700 1993 4,191,000 2,218,500 1,648,100 716,100 1994 4,116,000 2,110,800 1,605,600 778,800 1995 3,493,500 1,848,600 1,549,900 796,200 1996 3,260,000 1,740,400 1,444,600 729,900 1997 3,039,000 1,741,800 1,405,200 717,800 1998 2,776,800 1,587,900 1,319,800 675,900 Note: The serious violent crimes included are rape, robbery, aggravated assault, and homicide. Because of changes made to the victimization survey, data prior to 1992 are adjusted to make them comparable to data collected under the redesigned methodology. Estimates for 1993 to 1998 are based on collection year while earlier estimates are based on data year. The measures of violent crime come from two sources of data: 1) The National Crime Victimization Survey (NCVS) a household survey ongoing since 1972 that interviews about 80,000 persons age 12 and older in 43,000 households twice each year about their victimizations from crime. 2) The Uniform Crime Reports (UCR) that collects information on crimes and arrests reported by law enforcement authorities to the FBI. ---------------------------------------------------------------- Measures of Serious Violent Crime Source: http://www.ojp.usdoj.gov/bjs/glance/cv2.htm) (Footnote: The victimization data are from the National Crime Victimization Survey (NCVS), which began in 1972. In it a random sample of U.S. households is chosen, and household members age 12 and over are asked about their victimization experiences. The crime data are from CIUS, modified to be made comparable to the victimization data; that is, for robbery it means that all commercial robberies are excluded, as are rapes, robberies and aggravated assaults whose victims were under 12 years old. That the two sources of data are not converging means that the crimes that citizens say they have reported to the police are being recorded more completely by the police in their statistics. For a more complete description of the characteristics of the two crime measures, see The Nation's Two Crime Measures (BJS/FBI, 1995, http://www.ojp.usdoj.gov/bjs/pub/pdf/ntmc.pdf and in the recent issues of CIUS).) Coverage Gaps and Imputation However, in recent years more and more of the crime statistics reported by the police have not been based on crime counts but on imputed crime counts. Figure 3 shows how the percentage of the population covered by the UCR has changed over time; as can be seen, a long period of improvement in coverage has been followed in recent years by a reduction in coverage. (Figure 3. Percent of the U.S. Population Covered by the UCR Source: CIUS, 1953-97.) This rather considerable decline in population coverage in the 1990's is due in part to problems at the State level, in converting their crime reporting systems to comply with NIBRS requirements. As problems are dealt with, the coverage should return to above 95%. The missing coverage is not uniform over space and time. Most jurisdictions provide largely accurate crime reports every month while others do not, for reasons described later in the report. Since 1969 more and more States have passed legislation mandating the submission of crime data by local jurisdictions to state agencies, but very few incur penalties if they do not comply with such requirements. See Appendix C for a listing of the characteristics of State UCR reporting procedures. Appendix D shows the extent to which the 50 States have provided crime reports to the UCR since 1958, the year that national and State crime rates were first published. As can be seen from the patterns, some States have historically been able to provide close to 100 percent UCR coverage (and therefore low imputation rates). This response is true of about 20 States. The infusion of LEAA funding in the 1970's apparently permitted an additional 12 States to improve their UCR reporting systems. All experienced a reduction in the percent of crime imputed in the 1970's and continue to have low percentages of imputed crime. Some States, however, have experienced substantial problems in UCR reporting: *Complete data for Illinois, for example, have not been included in the UCR since 1985, initially because the Illinois statutory definition of sexual assault is inconsistent with the UCR definition of rape, (Footnote: "Until 1984, 'rape' was defined as the carnal knowledge of a female, forcibly and against her will. On July 1, 1984, Illinois' sexual assault laws became gender neutral and the old concept of rape was broadened to include many types of sexual assault. This index crime now includes all sexual assaults, completed and attempted, aggravated and non-aggravated." (Illinois Criminal Justice Information Authority, 1987, p. 5.)) and since 1992 because the Illinois UCR submissions did not adhere to the UCR's "hierarchy rule." *A number of States have had problems in implementing NIBRS, reflected in recent major increases in the percentage of imputed UCR data or in the complete absence of data. *In still other States, there has been a recent gradual growth in the percent imputed, reflecting a gradual withdrawal of local jurisdictions from the UCR reporting program. These reporting omissions (i.e., data that are missing or are reported too late to meet the publication deadline of CIUS) have generally been considered to be of little consequence, because most do not account for a significant percentage of overall crime -- as can be seen in figure 3, despite the reporting gaps depicted in Appendix D, the UCR still represents 87 percent of the U.S. population. Moreover, the FBI has developed procedures to accommodate such omissions. These procedures in essence "fill in the gaps" by imputing data when the data are either missing or not furnished to the FBI until after its publication deadline. Such imputations permit the FBI to make national, regional, and State estimates of crime data despite the missing data, and thus keep the annual publication of CIUS on schedule with relatively comparable data from year to year. ---------------------------------------------------------------- Figure 3. Percent of the U.S. Population Covered by the UCR Percent of U.S. population 1953 70% 1954 75 1955 80 1956 81 1957 83 1958 88 1959 88 1960 88 1961 92 1962 93 1963 92 1964 91 1965 92 1966 91 1967 92 1968 92 1969 92 1970 91 1971 92 1972 93 1973 93 1974 94 1975 95 1976 96 1977 98 1978 98 1979 98 1980 98 1981 97 1982 97 1983 97 1984 96 1985 97 1986 97 1987 96 1988 96 1989 96 1990 96 1991 96 1992 95 1993 90 1994 90 1995 90 1996 88 1997 87 Source: CIUS, 1953-97 ---------------------------------------------------------------- But researchers have been using county-level data to study crime characteristics, without realizing that some counties' crime statistics are based on a substantial amount of imputed data. (Footnote: Neither estimated city nor county data are disseminated outside the FBI. They are used solely to arrive at State and national estimates.) The county-level data set is compiled from the raw jurisdictional data provided by the FBI to the National Archive of Criminal Justice Data (NACJD), which uses its own county-level imputation procedures (described in Section IV.). NACJD is maintained by the University of Michigan's Inter-university Consortium for Political and Social Research (ICPSR). Funded by BJS, NACJD obtains the FBI's archived raw crime and arrest data sets to archive them on their own website in a form suitable for research use. NACJD has agency- level data files from 1966-96 and data files aggregated at the county level for 1977 through 1996. Imputation procedures used by NACJD in aggregating data to the county level are described in Section IV. As mentioned earlier, in some cases the data for a whole State have been problematic. In particular, over the past decade some or all of the data from Delaware (1995), Florida (1988, 1996), Illinois (1985-97), Iowa (1991), Kansas (1993-97) Kentucky (1988, 1996-97), Michigan (1993), Minnesota (1993), Montana (1994-97), New Hampshire (1997), Pennsylvania (1995), and Vermont (1997) have not been included by the FBI for tabulation in CIUS, as seen in figure 4. In other words, to develop national estimates of crime, data for States have been imputed in whole or in part. Imputation to such an extent may no longer be appropriate or desirable, especially now that UCR crime data are legislatively required to be used in formulas for allocating certain Federal funds. ------------------------------------------------------------------ Figure 4. State-Level Reporting of UCR Data to the FBI Indiana no State-level data Mississippi no State-level data Missouri no State-level data Utah no State-level data 1970-93 reported UCR data to the FBI 1994-97 South Dakota no State-level data 1970-92 reported UCR data to the FBI 1993-97 Dist. of Col. no State-level data 1970-80 reported UCR data to the FBI 1981-97 Washington no State-level data 1970-80 reported UCR data to the FBI 1981-97 North Dakota no State-level data 1970-78 reported UCR data to the FBI 1979-97 Connecticut no State-level data 1970-77 reported UCR data to the FBI 1978-97 Massachusetts no State-level data 1970-77 reported UCR data to the FBI 1978-97 New Hampshire no State-level data 1970-77 reported UCR data to the FBI 1978-96 had some problems with data 1997 Wyoming no State-level data 1970-77 reported UCR data to the FBI 1978-97 Alabama no State-level data 1970-76 reported UCR data to the FBI 1976-97 Colorado no State-level data 1970-76 reported UCR data to the FBI 1976-97 Ohio* no State-level data 1970-76, 1985-97 reported UCR data 1976-84 Texas no State-level data 1970-76 reported UCR data to the FBI 1976-97 Arizona no State-level data 1970-74 reported UCR data to the FBI 1976-97 Georgia no State-level data 1970-74 reported UCR data to the FBI 1975-97 Hawaii no State-level data 1970-74 reported UCR data to the FBI 1975-97 Iowa no State-level data 1970-74 reported UCR data to the FBI 1975-97 had some problems with data 1991 Kansas no State-level data 1970-74 reported UCR data to the FBI 1975-97 had some problems with data 1993-97 Louisiana no State-level data 1970-74 reported UCR data to the FBI 1975-97 Maryland no State-level data 1970-74 reported UCR data to the FBI 1975-97 New York no State-level data 1970-74 reported UCR data to the FBI 1975-97 Tennessee* no State-level data 1970-74, 1981-97 reported UCR data to the FBI 1975-80 Virginia no State-level data 1970-74 reported UCR data to the FBI 1975-97 Arkansas no State-level data 1970-73 reported UCR data to the FBI 1974-97 Idaho no State-level data 1970-73 reported UCR data to the FBI 1974-97 Illinois no State-level data 1970-73 reported UCR data to the FBI 1974-97 had some problems with data 1985-97 Maine no State-level data 1970-73 reported UCR data to the FBI 1974-97 Nevada no State-level data 1970-73 reported UCR data to the FBI 1974-97 New Mexico* no State-level data 1970-73, 1984-97 reported UCR data to the FBI 1974-83 Oklahoma no State-level data 1970-73 reported UCR data to the FBI 1974-97 Oregon no State-level data 1970-73 reported UCR data to the FBI 1974-97 South Carolina no State-level data 1970-73 reported UCR data to the FBI 1974-97 Michigan no State-level data 1970-72 reported UCR data to the FBI 1973-97 had some problems with data 1993 North Carolina no State-level data 1970-72 reported UCR data to the FBI 1973-97 Delaware no State-level data 1970-71 reported UCR data to the FBI 1972-97 had some problems with data 1995 West Virginia no State-level data 1970-71 reported UCR data to the FBI 1972-97 Florida no State-level data 1970 reported UCR data to the FBI 1971-97 had some problems with data 1988, 1996 Kentucky no State-level data 1970 reported UCR data to the FBI 1971-97 had some problems with data 1989, 1993-97 Minnesota no State-level data 1970 reported UCR data to the FBI 1971-97 had some problems with data 1993 Alaska reported UCR data to the FBI 1970-97 California reported UCR data to the FBI 1970-97 Montana reported UCR data to the FBI 1970-97 had some problems with data 1994-97 Nebraska reported UCR data to the FBI 1970-97 New Jersey reported UCR data to the FBI 1970-97 Pennsylvania reported UCR data to the FBI 1970-97 had some problems with data 1995 Rhode Island reported UCR data to the FBI 1970-97 Vermont reported UCR data to the FBI 1970-97 had some problems with data 1997 Wisconsin reported UCR data to the FBI 1970-97 Sources: CIUS, 1969-96, and responses from State officials by letter, e-mail, and telephone ------------------------------------------------------------------ The UCR and Funding Decisions In 1994, in reauthorizing the Omnibus Crime Control and Safe Streets Act of 1968, the U.S. Congress appropriated additional anticrime funding for jurisdictions under the Local Law Enforcement Block Grant Program. The amount of funds received by a jurisdiction was to be based on the number of violent crimes they had experienced in the 3 most recent years (1992-94). According to the statute, the UCR was to be the source of the crime data. This marked the first time that funding decisions were to be made on the basis of the data in the UCR, and caused a number of other agencies within the US Department of Justice (DOJ) to deal directly with the short- comings of this data set. The Bureau of Justice Assistance (BJA) was charged with the task of allocating the funds; BJA called on BJS, with its statistical expertise and know- ledge of the UCR's characteristics, to develop the allocation formula according to the law's provisions. [Appendix E gives the background for the development of this formula.] BJS used the actual raw crime data as reported by each police agency to the FBI, rather than the imputed data, in the allocation formula. But in reviewing the raw UCR data, BJS immediately recognized their limitations: Of the 18,413 police agencies that reported to the FBI in 1992-94, 3,516 (19%) did not provide crime data for any month during the 36-month period used in the formula and another 3,197 (17%) reported between 1 and 35 months (table 1). ---------------------------------------------------------------- Table 1. Reporting Behavior of 18,413 Police Agencies, 1992-94 Police agencies Reporting frequency Number Percent Total 18,413 100% Full reporting (36 months) 11,700 64% Partial reporting (1-35 months) 3,197 17 No reports 3,516 19 Special agency 2,650 14 Regular agency 866 5 ---------------------------------------------------------------- Although most gaps in the data were found to be relatively inconsequential, this was not true across the board. Of the 3,516 non-reporting agencies, all but 866 either were within jurisdictions that had other agencies report for them or were State agencies or special police agencies (such as transit police, fish-and-game police, or park police) that probably were not eligible for a formula award. (Footnote: But by not reporting crimes that occurred within the jurisdiction, they may have affected the statistics of agencies that were eligible for an award, and thus the crime figures reported to the FBI for that jurisdiction may be lower than had actually occurred. In some States an agency reporting as few as five violent crimes in the 3-year period qualified for a grant of over $10,000.) However, the remaining 866 agencies that provided no crime data for the 36-month period included some major jurisdictions: the primary police agencies in 3 cities and counties with populations over 100,000; 17 cities with populations between 50,000 and 100,000; and almost 200 cities with populations over 10,000. Note that fully 5 percent of the regular police agencies provided no reports for 3 full years. A subsequent analysis found that 15 percent of the regular agencies did not provide any data for 1992, and reporting behavior worsened in succeeding years (reanalysis by S. Lindgren, May 27, 1999). In less populous States, even cities with populations of 10,000 received awards. Thus, the lack of complete reporting had financial consequences for a significant number of jurisdictions. The legislation does make provision for determining funding if UCR data are not available, but it may also serve as a spur for ORIs to improve their reporting practices. The UCR and Electronic Access Another recent change in the crime data environment is the greater degree of public access to crime data. They have always been available to the public on paper, in the annual CIUS publications. For the most part, analyzing the data in the past usually meant entering data from the paper version of CIUS into one's own computer. (Footnote: When I started as a Visiting Fellow at BJS in early 1995, BJS statisticians were still doing this on a regular basis.) As discussed earlier, for many years they have been made available (primarily to researchers) in electronic form (e.g., magnetic tape), but they are now also accessible to the general public from various websites. The FBI, BJS, and NACJD now have regularly updated websites that provide access to UCR data, so it can be anticipated that more people will be encountering the inconsistencies in the data. Each site contains crime data, but in different forms and formats: *The FBI site is http://www.fbi.gov; it contains the UCR data as published in CIUS, beginning in 1995. *The BJS site is http://www.ojp. usdoj/bjs In the section Crime and Justice Electronic Data Abstracts (CJEDA), it provides UCR crime data by State from 1960 to 1997, UCR crime and arrest data for the 90 largest counties for 1990-96, and 1985-97 homicide data for cities with populations over 100,000. *The NACJD site is http://www.icpsr. umich.edu/nacjd/ucr.html; it contains downloadable arrest and offense data at the agency level from 1966 to 1996 and at the county level from 1977 to 1996. The State estimates provided by the FBI (and found on the BJS website) are based on the FBI's imputation and estimation procedures and are not directly comparable to NACJD's county-level data. Now that any person in the world with a computer and modem can download the data and do comparisons, it would be helpful to resolve the data inconsistencies as much as possible and to provide explanations for the inconsistencies when resolution is not possible. (Footnote: Many of the downloadable data sets currently contain explanations for some inconsistencies (see, for example, the data set at http://www.ojp.usdoj.gov/bjs/dtdata.htm#crime), but the explanations are not complete.) Use of Sub-National UCR Data The ready availability of UCR data at the subnational level has resulted in researchers using these data to answer policy questions. Unfortunately, the data may not be up to the task. To understand why this is so, a brief account of the history of their collection, aggregation, and initial uses would be beneficial. Prior to 1958, UCR data were collected from individual jurisdictions and aggregated to give, for each crime type in the crime index: urban and rural crime rates and year-to-year changes; crime counts by size of city (for reporting cities) and year-to-year changes; crime counts by State for reporting cities in each State, and year-to-year changes; and crime counts in cities by size of city, and year-to-year changes. State-level data were based on only those cities that reported to the FBI, and State-level crime rates were calculated by dividing the crime counts for these cities by their aggregate population. State-level data. Despite the known deficiencies in the data, the UCR State-level homicide data for the years 1930, 1940, 1950, 1960, and 1970 were used by Ehrlich (1975) to estimate that every execution deters eight homicides, a finding that the U.S. Supreme Court cited (Maltz, 1996, p. 36). Critics pointed out some of the analytic problems, but it was assumed that State-level homicide data would be more accurate than data concerning other crimes. However, the State-level homicide rates for 1930, 1940, and 1950 were doubtless based on jurisdictions covering less than 70 percent of the Nation. (See figure 3.) The variation in coverage from State to State was probably considerable. Although I have not examined the data from this era, it seems likely that much of the reporting was from urban agencies. Thus, a State that was 75 percent rural and 25 percent urban, but in which the urban agencies were much more diligent than rural agencies in reporting UCR data, would have its homicide rate based primarily on the homicide experience of its urban areas rather than on the experience of the State as a whole. County-level data. In 1983 BJS published Report to the Nation on Crime and Justice (Zawitz, 1983), a snapshot of the state of crime and justice at that time. It featured a choropleth map of the county-by-county violent crime rate in the United States in 1980 (Figure 5. Map: County-Level UCR Violent Crime Rates, 1980 Source: Zawitz, 1983). (Footnote: A choropleth map of crime displays levels of crime with different shadings or colors.) To produce this map, BJS tasked NACJD with estimating the crime rate of each county. This was done by aggregating the crime count for all the jurisdictions in the county and dividing by the aggregated population for those jurisdictions. Some counties reported no data; they are represented in white in the figure. ORIs that did not report at least 6 months of data to the FBI were also excluded; those that did report 6 months or more, but provided less than 12 months, had their data imputed. The imputation procedure simply multiplied the violent crime rate by 12/N, where N was the number of months reported. This implicitly assumes that the crime rate for non-reporting months is the same as for the reporting months. Moreover, if some agencies in a county did not report, or reported less than 6 months of data, their data and their population were excluded from the crime rate calculations. This implicitly assumes that the crime rate for nonreporting ORIs is the same as for the reporting ORIs in the county, which is probably a stretch. It should be noted that the imputation procedure was developed as an ad hoc procedure to make the 1980 data reasonably comparable from county to county so as to provide a snapshot, and not as a final means of dealing with missing data. However, this report was received so favorably that BJS decided to update it. In 1988 it released the second edition of Report to the Nation on Crime and Justice (Zawitz, 1988a), based on UCR data from 1984. (See Figure 6. Map: County-Level UCR Violent Crime Rates, 1984 Source: Zawitz, 1988a.) NACJD used the same imputation procedure to fill in the missing data. (Footnote: The procedure is described in Zawitz (1988b), p. 8.) Because of favorable reception of the reports and the data on which they were based, BJS decided to make county-level data sets routinely available through NACJD. The deficiencies or consequences of using the ad hoc imputation procedure were not considered, because up to that time the county-level data had only been used for cross-sectional comparisons and not for more rigorous analytic purposes. Since then, however, these data have been used for other purposes. For example, a recent study used the data to conclude that right-to-carry laws reduce crime (Lott, 1998). This finding was contested on methodological grounds (Black and Nagin, 1998), but not from the standpoint of the data quality. It turns out that smaller counties are more likely than the larger counties to have a significant fraction of their data imputed (C. Dunn, at the 1997 workshop); the fact that smaller counties are more rural may have a decided effect on this analysis. One data documentation feature that NACJD now uses (until an improved imputation procedure is implemented) is a "coverage factor" in the county-level data set. This feature (described in Section VI) at least warns the analyst that the data are limited in coverage. The UCR and NIBRS A fourth change in the crime data picture concerns the way crime data are to be reported to the FBI. Over 10 years ago a study commissioned by the FBI and BJS provided a "blueprint" for changing the way crime data were to be reported to the FBI (Poggio et al., 1985). The recommended changes have been adapted and incorporated in a set of new procedures that comprise NIBRS; NIBRS has already been implemented in a number of states and is expanding to cover the entire United States. The change in data collection is considerable. Under the UCR program an agency provides a monthly summary report of crime, called Return A (Figure 7. Replica of the FBI's UCR Return A); each line of the report refers to a single type of crime; it contains a count of the number of crimes of that type that had occurred in that month. Under NIBRS each incident is to be reported in detail, with a number of records devoted to describing the characteristics of each crime. For a single incident, information is recorded for each included offense (type, weapons, location, motivation method of entry, etc.); victim, offender, and arrestee; type of property; and so on. See figure 8 (Figure 8. Diagram: The Structure of NIBRS Data Elements Source: Akiyama and Nolan, 1999a). NIBRS will provide a great deal of detail about the nature of criminal activity: for example, one will be able to determine to what extent aggravated assaults were committed by family members or strangers, or what fraction of burglaries occurred in apartments or in private homes, by time of day, and in other ways. This will give both the police and the public with detailed information on the risk of crime to enable them to develop more useful policies and tactics. One of the problems with current UCR reporting that should be ameliorated by NIBRS is caused by a characteristic of the UCR system known as the hierarchy rule. The hierarchy rule for reporting crimes was instituted by the FBI in the 1930's, to ensure that there would be no double-counting of crimes. A criminal event that includes two different crime categories is thus counted only once, and only in the most serious crime category. For example, if a convenience store robbery results in the death of the store clerk, this would be classified as a homicide rather than a robbery -- because homicide is a more serious crime than robbery. Yet this expedient, important in the pre-computer age, masks the nature of what happened. It would certainly be better to recognize both characteristics of the incident, if only to be able to provide an estimate of risk, in the form of the fraction of incidents that start out as robberies but result in homicide (see, e.g., Maltz, 1976b). Even with NIBRS implemented, summary data will doubtless be aggregated and compiled for each agency, and data for some agencies may continue to be missing, delinquent, or in error. In other words, there will still be a need for imputation procedures after NIBRS is implemented nationwide. In fact, missing data may become a greater problem under NIBRS, because of the huge increase in categories and the complexity of definitions. This may make it more difficult to assume that the counting rules and definitions are being applied uniformly. III. Incomplete Crime Data Two separate streams of crime data are sent to the FBI's Uniform Crime Reporting Section: one from individual police agencies, the other from State UCR collection programs. Although 36 States now have statutes mandating the reporting of crime and other criminal justice information, not all police departments submit this information to their State agency designated to collect the data, or they may submit it too late for entry in CIUS. Occasionally some of the data may appear to be in error -- too high or too low, based on the jurisdiction's past crime experience. This section describes the procedures used by the FBI to correct errors in reporting and, when reporting gaps occur, to impute data as necessary. It also discusses when and how UCR data files are updated. Error Checking Potential errors in the data are checked in different ways, depending on how the UCR reports were sent to the FBI and depending on the size of the jurisdiction. If the data are first collected by the State agency, that agency itself may undertake follow-up procedures to verify the data. When the errors are glaring, they can be found by simply inspecting the data or by using simple graphical techniques. One State agency refers to such errors as "tent poles" and "craters" -- excessively high or low figures compared to the surrounding data (R. Christ, personal communication). Such errors often come from transposing numbers in returns submitted manually. Small errors, however, will probably not be caught in this manner. If the State does not have auditing procedures, or if the data are sent directly to the FBI, staff members in the FBI's UCR Section may note the omission or anomaly and request the State agency to follow up. In cases in which the data are sent directly to the FBI, the FBI may follow up with the police agency by mail. If, however, the agency has a population of over 100,000, personnel from the UCR Section call the agency directly to verify the data (D. Kording, at the 1997 workshop). When errors are found in the data, they are corrected, and the corrected counts are included in the statistics. Depending on when the errors were discovered and corrected, they may not be incorporated in CIUS (if the corrections occur after the FBI's publication deadline), but they may be included in the public-use data set archived at NACJD (J. Lynch, at the 1997 workshop). This means that someone trying to determine the extent of crime in a jurisdiction will encounter unexplained differences between CIUS statistics and the data archived by NACJD. Reasons for Incomplete Reporting Aside from these errors in reporting, police agencies may not provide complete (or any) reports to the FBI. The agencies may be delinquent or incomplete in their reporting of crime for a number of reasons: *Some agencies experienced natural disasters that prevented them from getting their data in on time (or in some instances, at all). *As has been the case with other public agencies, budgetary restrictions on the police have meant that some agencies have had to cut back on services. Although crime reporting is considered an essential function because it provides information about community safety to the public, some agencies that are especially strapped may forgo these routine clerical activities so as to ensure that sufficient resources exist for patrolling the streets. *Retirements, promotions, and other personnel changes may mean that the person experienced in the preparation of UCR crime and arrest data is replaced by someone -- -- who has little experience in its preparation (and consequently makes numerous errors) -- who is not given sufficient training -- who gives the task a low priority -- or who doesn't prepare the data in a timely manner. *With respect to training, some jurisdictions may rely completely on handbooks on UCR reporting produced by the Program Support Section, and there may be ambiguities in the reports that require more complete descriptions than are included in the handbooks. *Phasing in a new reporting system or computerization of the old system may cause delays or gaps in the crime reporting process. This may be especially true as agencies convert to NIBRS. (See Appendix D.) *Small agencies with little crime to report may feel it unnecessary to fill out reports that are filled almost entirely with zeros. [In fact, in some cases small agencies file reports for only 1 month; they want to ensure that their agencies' employee statistics are included in CIUS, and reporting their data for 1 month will accomplish this.] *A State may have offense definitions that are incompatible with UCR definitions, leading to data being submitted but not accepted. Thus, there are a number of reasons that crime reports may be incomplete, late, or in error. The extent to which this is a problem in an individual State can be seen in Appendix D (page 47), which shows the UCR reporting behavior of each State over the past 40 years. Note that the impact of LEAA funding of State statistical systems in the 1970's is apparent in these graphs, as is (in some States) the impact of its termination. Note also that while some States have a history of consistently good reporting, other States have a history of consistently poor reporting, and yet others have exhibited highly erratic reporting behavior. In particular: *The data for six States were excluded from the 1997 UCR, with the data from one of those States not having been included since 1993. *Six States have consistently poor reporting, missing reports on the crime experienced by more than 20% of their population. Some of the recent erratic reporting by States is attributable to their conversion to NIBRS. In particular, some States and agencies that have begun the NIBRS conversion process are working with software that currently does not have the ability to produce UCR reports. Over the long term we can expect that many of these reporting problems will disappear or at least diminish. Many smaller agencies that are currently automating are purchasing computer software that provides near-automatic reporting (including audit checks) of these data. However, in the near term we can expect these problems to continue, for standards for such software do not currently exist. (See Appendix C.) Incomplete Arrest Data As with offense data, there are major gaps in arrest data. To some extent the problems with arrest data are greater because of three factors: *The percent of arrests reported by police is substantially lower than the percentage of crimes reported. *By publishing the characteristics of arrestees, there is an implicit assumption that they also characterize those who commit similar crimes but are not arrested. *Whereas crimes reported to the police are generally considered to be (and have been shown to be) similar to crimes not reported to the police, arrests reported by the police are as much a reflection of police priorities as they are of criminal activity. Agencies are less diligent in reporting arrest data than crime data. The FBI attempts to ensure completeness of arrest data by rejecting an agency's adult arrest data if it does not also send in juvenile arrest data, nor is arrestee race information accepted without age and sex information (V. Major, at the 1997 workshop). This means, however, that information on arrests is considerably less complete than information on crimes. Moreover, the arrest data published in CIUS are biased even in comparison to the arrest data eventually reported to the FBI. (Footnote: The FBI accepts data that it receives after its publication deadline date and includes them in data files sent to NACJD, so the data files include reports not included in CIUS.) For example, Snyder compared the 1980 CIUS arrest data with the final counts of arrests, after all of the late-reporting ORIs submitted their data (H. Snyder, personal communication, 1999). He found that juvenile arrests (as a percentage of all arrests) were overrepresented in the published statistics. He attributed this to the fact that large urban agencies, with higher percentages of juvenile arrestees, generally reported early (in time for publication) and the late reporters tended to be less urban agencies, with lower percent- ages of juvenile arrestees. So, not only are the arrest data published in CIUS not a representative sample of all arrests, they are not a representative sample of arrest data eventually reported to the FBI. Figure 9 shows how 1997 arrest reporting varies by State. As can be seen, 4 States and Washington, DC, did not provide acceptable arrest data, and more than half of the population was not represented in an additional 12 States. --------------------------------------------------------------- Figure 9. Percent of Population Covered in Arrest Data Reported to the FBI, 1997 Percent of population State covered by reported arrest data District of Columbia 0% Florida 0 Kansas 0 New Hampshire 0 Vermont 0 Kentucky 19 Illinois 23 Georgia 33 Mississippi 37 Montana 39 Delaware 41 Tennessee 42 Alaska 44 South Dakota 45 Nevada 46 New York 46 Pennsylvania 47 Ohio 55 Indiana 57 Missouri 60 Washington 61 New Mexico 68 Colorado 70 Utah 75 Wisconsin 76 Louisiana 79 Michigan 81 Iowa 82 Massachusetts 83 Arizona 85 Connecticut 85 Oregon 87 Arkansas 89 North Dakota 90 Alabama 94 Maine 94 Nebraska 94 West Virginia 96 New Jersey 96 Hawaii 97 Texas 97 Idaho 98 Wyoming 98 Virginia 98 California 99 North Carolina 99 South Carolina 99 Minnesota 100 Maryland 100 Oklahoma 100 Rhode Island 100 Source: CIUS, 1997 --------------------------------------------------------------- Figure 10 shows the extent nationally to which arrest data have been reported to the FBI for the past four decades. The percent- ages are consistently lower than those for crime data (cover figure), but their time trend shows the same declining pattern. As with the decline in crime reporting, it is probably attributable for the most part to the difficulties in shifting to NIBRS. (Figure 10. Percent of U.S. Population Represented by Agencies that Provide Arrest Data Source: CIUS, 1960-97) --------------------------------------------------------------- Figure 10. Percent of U.S. Population Represented by Agencies that Provide Arrest Data Percent of U.S. population 1960 61% 1961 63% 1962 67% 1963 68% 1964 70% 1965 69% 1966 70% 1967 73% 1968 73% 1969 72% 1970 75% 1971 76% 1972 77% 1973 74% 1974 64% 1975 84% 1976 82% 1977 88% 1978 95% 1979 93% 1980 92% 1981 94% 1982 81% 1983 86% 1984 76% 1985 85% 1986 82% 1987 83% 1988 77% 1989 81% 1990 78% 1991 75% 1992 84% 1993 83% 1994 80% 1995 74% 1996 72% 1997 68% Source: CIUS, 1960-97 --------------------------------------------------------------- Note that compared to the reporting of crime data, there is a greater degree of annual variation in the reporting of arrest data. Year-to-year differences of close to ten percent in the reporting population are not uncommon. This variation is due in part to the changes in reporting standards for arrests. For example the large "notch" in arrest reporting in 1974 was probably due to the changeover from annual to monthly arrest reporting, which took some time for agencies to systematize (V. Major, personal communication, May 20 and August 23, 1999). In prior years only agencies that reported arrests every month (i.e., were "12-months complete") were included in the arrest tallies; starting in 1974, when monthly arrest data began to be collected, arrest data were aggregated for all agencies with 6 months or more of arrest data. This changed again after 1981; from 1982 on, the FBI reverted to reporting aggregate arrests for only those agencies that provided 12 months of arrest data. The effect of this change can be seen in the 1981-82 drop in the population represented in arrest reports. This lack of completeness and consistency (and, more importantly, lack of representativeness) of the reporting of arrest data can have major consequences because of implicit assumptions made by some individuals in "analyzing" the arrest data. For example, Snyder shows how these arrest data have been used improperly to infer offense rates of juveniles (Snyder, 1999). V. Processing and Publishing the Crime Data After the data are received by the FBI (specifically, by the Program Support Section of the Criminal Justice Information Services Division), they are stored in a data file that contains (among other data) offense data for each ORI taken from Return A (figure 7): month-by-month counts of each of the offenses listed. A computer program processes this file, in which the 12 months of data for each offense are summarized by two numbers: total for that offense and number of months reported. A page of output from one of the many programs used by the FBI to process the data is shown in Figure 11. (Figure 11. Part of the Printout of the FBI's Crime-by-County File ) This output file, "Crime by County," is one of then more widely distributed files. Note that -- *The ORIs are grouped by State and by county within each State. *An ORI's population and its Standard Metropolitan Area (SMA -- not the same codes as used by the U.S. Bureau of the Census) and population group indicators are given. (See table 2.) ---------------------------------------------------------------- Table 2. FBI Classification of Population Groups Population group Political label Population range I City 250,000 and over II City 100,000 to 249,999 III City 50,000 to 99,999 IV City 25,000 to 49,999 V City 10,000 to 24,999 VI City/a Less than 10,000 VIII (Rural County) County/b . . . IX (Suburban County County/b . . . Note: Group VII, missing from this table, consists of cities with population under 2,500 and universities and colleges to which no population is attributed For compilation of CIUS, Group VII is included in Group VI. a/Includes universities and colleges to which no population is attributed. b/Includes State police to which no population is attributed. ---------------------------------------------------------------- *Both the total number of Index crimes and modified Index crimes (the first seven Index crimes plus arson) are given. (Footnote: The arson totals provided in figure 11 are summaries of all arson on file for the respective agencies and may not be representative of the number of months reported in the MO column.) *Although not apparent from the printout, only the reporting ORIs are included in this compilation. If an ORI does not submit any UCR data for a year, it is omitted from the data file, and its population is not included in the county total. Publishing the UCR The deadline for submitting UCR data to the FBI is late March of the following year. Data submitted beyond this date are accepted by the FBI and incorporated in the data file until the FBI closes out the data file for that year. The date that this file is closed out is not fixed; for example, the 1997 file was not closed out until early April 1999. The paper version of CIUS for a given year is published in the fall of the following year, usually in October or November. Between the March cutoff date and the publication date the staff of the Program Support Section perform the error checks and prepare the data for publication. January and February are devoted to writing to the larger ORIs to verify data and/or to request missing months; listings of missing months are routinely prepared during this period. Data from agencies that do not provide 12 months of data are analyzed to identify any month(s) deviating from agency norms due to special circumstances affecting those agencies (e.g., floods, tornadoes, and fire). By early March all agencies with delinquent data have been contacted. By mid-February population estimates are calculated, based on Census Bureau data, and included in the raw data file. By mid-April the data processing unit prepares a preliminary set of tables; this permits the Crime Analysis Research and Development Unit to begin to look for patterns in the data and draft the text and analysis for the report. In addition, the tables are sent to the outside contractor to format the tables for printing. The material for CIUS is sent to an contractor in three installments. The first installment is delivered to the contractor in May. It consists chiefly of the appendixes, which have few tables and do not change much from year to year, the methodology section, and tables of law enforcement personnel, which had been collected from the ORIs in October. Over the next month the FBI corrects the proofs and returns them to the contractor twice in succession. Installment 2, consisting chiefly of Crime Index Offenses Cleared, is sent to the contractor in early June and, as with the first installment, is proofread and returned to the contractor twice for revision over the next month. The third installment consists primarily of tables and text on offenses and arrests, as well as the program summary (Section I) and the inclusion of some data that were omitted from sections that had been prepared earlier. For example, the schedule for the 1998 CIUS projects completion of these three installments by the end of July, and a final check of the entire report by early August. Archiving the UCR Data File At the request of BJS, the raw data file is provided to NACJD for archiving. The file is then restructured by NACJD and additional fields are included to make it more accessible for research purposes. In the past this restructuring has resulted in errors such as mismatched fields; consequently, the FBI cannot respond to queries about the data archived by NACJD. The file that NACJD archives is not the one used to produce CIUS; rather, it is the updated file that contains the additional data received by the FBI after the March publication deadline. This has meant that analyses using the raw data cannot be compared to the tables in CIUS, because they are based on different data sets. NACJD also produces county-level files from the raw data file (see Section II) which are available for downloading from NACJD (http://www.icpsr.umich.edu/nacjd/ucr.html) . Notes accompanying those files state that "UCR county-level files are not official FBI UCR releases and are being provided for research purposes only. Users with questions regarding these UCR county-level data files can contact the National Archive of Criminal Justice Data at ICPSR." Thus, two sets of UCR data are made available to the public. One, published in CIUS and available through the FBI website, contains the data sent to the FBI before its publication cutoff date. The other, available through the NACJD website, contains data sent to the FBI before its data file cutoff date, which may be considerably later than the publication cutoff date. The difference between the two is usually not great, but has led to some misunderstandings. VI. Procedures Used for Imputing Crime Data It should be noted at the outset that the FBI does not publish or release data that include imputations below the State level. Its imputation procedures are used solely for estimating crime rates at the State and national levels. NACJD, however, does publish offense data that have been imputed at the county level. In this section we describe the imputation procedures used by both the FBI and NACJD. FBI Imputation Procedures for Crime Whether an ORI reports through the Summary UCR Program or through NIBRS, the FBI will still need to use imputation techniques that allow it to make reasonable estimates of crime and arrests. The imputation procedures used by the FBI for estimating crime rates are described below. Since 1958 the FBI has used two different means of imputing crime data for a police agency. (Footnote: Crime data are not imputed for all agencies; see "Imputation and Zero-Population Agencies.") If the agency reports 3 or more months, one procedure is used, while another is used when less than 3 months is reported. Partial Use of Data. If an agency has provided reports of crime data for 3 or more months, the imputation procedure is based on those reports. The total annual crime for that jurisdiction is estimated by multiplying the reported number of crimes by 12/N, where N is the number of months for which reports exist. Thus, an agency that reports 4 months of crime data (a third of the year) would be estimated to have 12/4, or 3 times the number of crimes that it reports for that period. Data Not Used. If an agency reports for 2 or fewer months, the number of crimes is estimated from scratch. These agencies are considered to be nonreporting agencies, and the FBI bases the imputed data for such agencies on the crime rates for the same year for similar agencies. "Similar agencies" are considered to be those in the same Population Group in the same State, but only those that provided 12 months of data. Table 2 (page 21) shows how the FBI categorizes these groups. Thus, if an agency in Alabama with a population of 150,000 reports 2 months or less of crime data in 1997, and the 1997 aggravated assault rate for Group II agencies (population between 100,000 and 249,999) in Alabama is 620.2 per 100,000, then the agency is estimated to have had 930.3 (620.2 x 150,000 / 100,000) aggravated assaults for 1997. (Footnote: If there are no comparable ORIs in the State, the estimate is based on the rates of occurrence in the region: New England, Middle Atlantic, East North Central, etc.) NACJD Imputation Procedure Every year NACJD obtains a data set from the FBI containing the raw UCR figures from the FBI, archives it, and uses it to develop a file containing crime and arrest data for each county in the US. NACJD also has to contend with missing data, in aggregating to the county level. It has used two different imputation procedures: one for the 1980-93 data sets and the other for datasets from 1994 onward. As stated earlier, the original county-level imputation procedure was developed to be used to plot crime by county for a single year, 1980. When BJS decided to continue providing county-level data through NACJD, they continued to use the same imputation procedure, similar to the FBI procedure but with a different cut point: agencies that provided reports for 6 or more months were estimated to have 12/N crimes, where N is the number of months reported. Those reporting less than 6 months were estimated to have the same offense rates as the rest of the county (not State and population group). If an agency with 10 percent of a county's population provided reports for 5 months or less, then the county's rates were used for that agency, and the estimated number of crimes for the county became C/.9, where C is the number of crimes reported by the rest of the county. However, from the 1994 data file onward NACJD has used essentially the same imputation procedure as the FBI. Coverage Factor. More recently, in consultation with BJS, NACJD began to include a new element, "coverage factor," in its county-level data files. This factor represents the extent to which the crime and arrest figures for a county are based on real data and the extent to which they are based on imputed figures. For example, if a county with a population of 500,000 includes an agency with a population of 100,000 that reported for only 9 months (all other agencies reporting fully), then the coverage factor for that county would be 1.00 - (3/12) x (100,000/500,000) = .95, or 95 percent, because a fourth of the data are missing for 20 percent of the county population. If that agency reported for 2 months or less, then the coverage factor would be 1.00 - (12/12) x (100,000/500,000) = .80, or 80 percent, since all 12 months are considered missing. This does not correct the problems of imputation so much as it puts the users of the data on notice that the data they are using have been estimated to some degree. Imputation Procedures for Arrests Arrest data are missing to a much greater extent than crime data. Imputing the missing arrest figures thus becomes much more difficult. The imputation procedure used by the FBI to account for missing or late arrest data is applied to all ORIs that report fewer than 12 months of data, i.e., are not "12-months complete." It is similar to that used for crime data, with two exceptions: first, arrests are estimated only at the national level; and second, instead of basing the imputation on the arrest rates in the same population group and State, it is based on the arrest rates in the same population group throughout the country. NACJD has used a similar arrest imputation procedure since the 1994 data year, except that data for the nonreporting agencies (2 or fewer months) are imputed based on the arrests rates in the same population group and the same State. Imputation and "Zero-Population" Agencies In compiling its crime and arrest statistics, the FBI tries to ensure that both the numerators (number of crimes and arrests) and denominators (number of people) are based on accurate data and estimates. This means that populations that are policed by more than one agency should be counted only once. Some jurisdictions are policed by State or county police departments, or even by other cities. For example, Chicago is in Cook County, Illinois. But the number of crimes reported by the Cook County Sheriff's Police Department (CCSPD) is for the areas policed by the CCSPD, and only for those areas (both unincorporated areas and municipalities that contract with CCSPD); it does not include the crimes in Chicago or any other Cook County jurisdiction not policed by the CCSPD. Since the crime rate is the number of crimes divided by the population, the population in question should live only in the areas policed by the CCSPD. As noted earlier (and in the footnotes to table 2), not all police agencies have populations associated with them in the UCR program. These "zero-population" agencies (transit police, park police, university police, and similar agencies) may be entirely within "primary" police jurisdictions (Groups I-VI in table 2) that already report crime to the FBI. Were these special police agencies to be associated in the UCR with the populations they serve, it would be tantamount to double-counting those populations. When these agencies report crime data, the data are attributed to that ORI and are included in the respective counties of the agencies. When they are not reported (or are delayed in reporting, or report only partially), no imputation is made of the missing figures. State police agencies are usually considered "zero-population" agencies because, for the most part, they police State highways and rural areas not covered by municipal agencies. When these agencies report crime or arrest data, the data are also attributed to their ORI (for example, each State police barracks may have its own ORI) and are included in the respective counties of the agencies. And similarly, when these agencies' data are not reported or are delayed in reporting or are reported only partially, no imputation is made of the missing figures. Updating UCR Files As previously noted, UCR files are updated when additional or corrected information is made available . For example, suppose a jurisdiction was missing December's Return A, but then sent it in after the publication cutoff date and prior to the date that the data file is archived. In such a case the data file (but not the published material) is updated to include the additional information. The updated file is then sent to NACJD for archiving, and contains different data than the printed version. VII. Inaccuracies Produced by the Imputation Procedures As is the case with all estimates, those produced by the FBI's imputation procedures are not entirely accurate. It may actually be that the inaccuracies due to imputation do not amount to much, at least for some uses of the crime data. But currently we have no way of knowing whether they produce major or minor discrepancies in the crime data, because the extent to which this imprecision affects our estimate of the crime rate has not been determined. Most observers believe that the effect on the estimate of the overall crime rate in the United States would be minimal, but that it could be quite problematic when investigating the crime rate for a smaller unit such as a State or county, or when looking at rural crime rates. The potential effects of the imputation procedures are described in this section. They include those used for primary police agencies reporting between 3 and 11 months (called "incomplete-reporting agencies"), for agencies that report less than 3 months (called "nonreporting agencies"), and for police agencies with no associated population (called "zero-population" agencies). Incomplete-Reporting Agencies Included in this category are agencies that report between 3 and 11 months of data, for which full-year estimates are made by inflating the data to cover the entire year. Even when this imputation procedure is used, biases may exist. This would be especially true for crimes that vary seasonally. If the months that were not reported are historically lower in crime, then basing the jurisdiction's annual crime rate on the remaining months would result in an overestimate. For example, an agency in a resort area may report its crime only during the tourist season (when there are more officers and more crime), so the imputed crime rate might be considerably higher than the actual crime rate. Nonreporting Agencies Included in this category are agencies that report data for 2 or fewer months, since their data are not incorporated in the agency's estimated crime rates. As described earlier, imputation of the crime rates of these agencies is based on other agencies in the same population group and State. However, it may be that the nonreporting agencies do not report specifically because they have little crime to report. If this is the case, then this procedure may overestimate the amount of crime. For the most part, the nonreporting agencies are in jurisdictions with quite small populations; however, in some counties the effect of such imputation may increase the estimated crime rate considerably. "Zero-Population" Agencies Included in this category are those agencies that have statewide jurisdiction (e.g., State police, fish and game police), or that are entirely within jurisdictions policed by agencies that already provide crime reports (e.g., university police), or are cities with populations under 2,500. When these "zero-population" agencies do not report crimes or arrests, no imputation is made of their estimated crime. For example, crimes and arrests made by the University of Michigan Police are reported to the FBI for UCR purposes. However, if the University of Michigan Police neglect to report the data or are late in getting the data to the State, no imputation is made of the missing data. This omission may distort the picture of crime and arrest volume, depending on the amount of crime that is not reported. For another example, suppose a State police agency's crime statistics are not produced in time for the FBI's publication cutoff date for CIUS. Suppose further that in one (rural) county the State police barracks is responsible for half of the crime reports. Since no provision is made for their imputation, the crime figures for that county would be underestimated by about 50 percent. In other words, the effect of not imputing zero-population agencies may not have much effect in terms of national or State-level statistics, but it can have a measurable effect on city- or county-level statistics. Summary In 1958, when the FBI first began to provide national estimates -- based on the recommendations of its Consultant Committee on Uniform Crime Reporting (FBI, 1958) -- about half of the States had over 90 percent coverage, and now about three-quarters have over 90 percent coverage. Yet, despite the increases in coverage, the missing data can distort the crime picture in subnational estimates. When a primary police agency (i.e., in Groups I-V in table 2) does not report, or provides partial reports, an attempt is made to rectify the omissions by imputation. The imputation procedures that are used, however, may serve to overestimate the amount of crime that actually occurred. When reports from a zero-population agency are missing, no matter how large the agency or how important it is in terms of crime reporting, no estimates are made to compensate for the missing reports. Thus, the effect of this policy may be to underestimate crime and arrest rates for counties and rural areas. The FBI initiated these procedures over four decades ago, well before the time when computing was widespread, and when compilation of crime and arrest data from each individual agency was a difficult and time-consuming process. Although it is still difficult and time consuming, the current state of computer hardware and software makes it possible to make adjustments to the data with much greater ease. That this imputation procedure has continued to the present is probably attributable to -- (a) the desire on the part of the FBI to maintain consistency in its data series (b) the assumption that it made little difference at the national level (at most 1 or 2 percent) and would not greatly affect the general trends in the data (c) the fact that there had been no need in the past to make any changes -- "If it ain't broke (or if no one cares if it's broke), don't fix it." But now that UCR data are being used at the jurisdictional level to determine funding, it is clear that the crime reporting system needs to be improved. (Footnote: Data for individual jurisdictions are not imputed in the calculation of funds; "however, the State-level estimates we use to determine the amount to be distributed across the States are based on imputations. To the extent that these imputations cause errors that are not consistent across all States, it does affect the dollar amounts to be distributed across the States" (S. Lindgren, memorandum to the author, March 20, 1999). Therefore, the agencies within States that have higher imputed estimates of crime than they actually experienced will get larger awards than they should, and those in States with too low estimates will get lower awards.) According to NACJD personnel, even before the new Federal legislation brought the problems to the forefront, it was quite evident that the crime rates in some counties had substantial discrepancies due to missing data (C. Dunn, at the 1997 workshop). Even though the FBI does not impute at the county level (and does not condone such uses), it is a fact that the data are imputed at this level -- and that policies are proposed based on the imputed data. It is for this reason that the FBI should consider revising its imputation procedures. A small percentage difference in overall crime is one thing, but when looking at a level like rural crime in a single State (or more specifically, like a single jurisdiction's crime rate), the difference might be more substantial. VIII. Suggested Imputation Philosophy [This section is based on conversations with and memoranda written by Yoshio Akiyama and James Nolan, III. While I am deeply indebted to them for their advice and consultation, as with the rest of this report the opinions and recommendations are mine and should not be taken as policy of the FBI or BJS.] The UCR imputation procedures used by the FBI have been generally effective. The goal of the FBI has been to estimate national and State rates, and the methods perform that task effectively. The extension of imputation to smaller units of analysis (where problems arise) was first done by NACJD at the behest of BJS, to provide a more fine-grained national picture of crime, and has since been done on a yearly basis by NACJD. This may have led researchers to (incorrectly) assume that the imputed data were accurate at this level. However, because the UCR data are the only source of crime and arrest data at the jurisdictional level and because they have been used to allocate Federal funds, it may well be worthwhile to update the imputation procedures to an extent. This will permit the FBI to provide as accurate an estimate of jurisdictional data as possible; however, it should be accomplished without unduly burdening the FBI with complicated procedures. No specific imputation procedures are recommended. Only those who have a detailed knowledge of the data collection and analysis processes can specify what makes sense in practice. However, some general principles can be considered. Suggested Imputation Philosophy At the jurisdiction level, a longitudinal estimation procedure (i.e., over time within the same jurisdiction) appears to be preferable to a cross-sectional one (i.e., for the same year across jurisdictions). Longitudinal estimation assumes that the best indicator of a jurisdiction's current and future crime and arrest activity is its own crime and arrest history, not the history of "similar" jurisdictions. This means that an estimate based on the jurisdiction's crime experience in the previous year is better than an estimate based on other jurisdictions' crime experience during the current year, and an estimate for a missing month that is based on the same month last year is better than one based on the reported months for this year. Current research in criminology strongly supports this approach. Recent studies have clearly shown that "all crime is local" (Sherman et al, 1997; Lattimore et al, 1997). That is, different jurisdictions have different crime patterns (and different neighborhoods in a jurisdiction have different crime patterns). This should be recognized in terms of how gaps in criminal justice data are filled in by imputation. For global estimates of missing agencies, however, cross-sectional methods should be seriously considered. They assume that the best indicator of a group of agencies is the average of similar agencies for the same year, as sampling theory indicates. Among the points that must be considered in developing an imputation scheme are (a) whether an agency reported data in the previous year (to permit longitudinal estimation), (b) the historical reporting behavior of a delinquent agency, (c) whether non-reporting agencies constitute a random sample of all jurisdictions, (d) whether the agencies in question have changed their borders during the time period in question. Different procedures need to be considered for zero-population agencies and for regular agencies that are incomplete- and non-reporting agencies. Some of the issues that might affect the imputation procedures are described below. Zero-Population Agencies Currently no imputation is made of missing data from zero-population agencies. This policy should be reconsidered. Although the crime counts from these agencies may constitute a small fraction of the whole, it may be advisable to investigate the extent to which ignoring these agencies' data affects the estimates of crime rates for specific sub-populations -- rural areas, certain types of crimes, etc. In particular, before any specific imputation procedure can be recommended, a study should be undertaken to determine (a) if different types of zero-population agencies should be handled differently and (b) whether they provide reports to other governmental entities (cities, States) from which an estimate of their statistics can be made. The examples given earlier, of university or State police agencies not providing data consistently, indicate that they may have some effect on biasing the crime picture; it is worth determining the extent to which this is the case. Different strategies might be tried for different situations. Imputing data for a non- reporting State police department might be based on the crime rate for the State as a whole; if it has decreased by 5 percent, then a reasonable estimate for the State police might also be a decrease of 5 percent. Imputing data for a university police department, on the other hand, might be based more properly on the change in crime rate for the city in which the university is located. [Although it might be more accurate to base the imputation on the change in crime rate for the neighborhood of the university, such detail is beyond the capability or needs of the UCR program. Incomplete and Non-Reporting Agencies As Akiyama and Nolan (1999b) point out in their memorandum, "UCR Data Imputation: Longitudinal vs Cross-Sectional Approaches," the bulk of the delinquent agencies report fewer than 4 months (i.e., there are relatively few incomplete-reporting agencies as compared to non-reporting agencies). Imputing the missing data for these agencies might best be accomplished by imputing each missing month separately. This has the effect of simplifying the imputation process by putting all agencies with missing data into the same category (and eliminating the arbitrary 3-month cut-point), but it also would permit seasonality to be taken into account by providing monthly estimates. There are two different cases to consider for primary police agencies that submit no reports. First, there are agencies that virtually never provide reports of crimes. The voluntary nature of the UCR program precludes the FBI from taking any measures to require reporting from non-reporting agencies. The fact that they do not report to the FBI, however, does not mean that they provide no reports of their activity at all; most agencies must do so, to some governmental body. A sample of these reports could be analyzed to determine the extent to which the current imputation procedure (i.e., using the same crime rate as in "similar" jurisdictions) reflects their true crime rates. For example, if a particular municipal police department does not provide crime data to the FBI or State criminal justice agency, it still may prepare an annual report to the municipality: a police department must normally justify its annual budget with an accounting of its activity. To determine the extent to which crime is underreported, it might be worthwhile to get in contact with a random sample of non-reporting agencies and obtain their annual reports. In this way an estimate could be made of the extent of crime and arrest activity in these and similar non-reporting jurisdictions. Second, there are agencies that normally do prepare reports but are unable to do so for a particular year. In such a case, an imputation procedure that is partly longitudinal and partly cross-sectional may be useful. That is, one can use the year-to-year change in crime rate for like jurisdictions (in the same state and same population group), and apply this year-to-year change to the agency's prior year's data. For example, if Agency X did not produce reports for 1997, and the year-to-year increase in crime for "like" agencies (i.e., in the same population group and State) was 3%, this increase could be imputed to last year's data for Agency X. IX. Supplementary Homicide Reports The FBI initiated the Supplementary Homicide Reports in the 1960s (Riedel, 1990), and NACJD has archived the data from 1976 to the present. All police departments that report homicides to the FBI generally also submit SHR forms. While the monthly reports to the UCR program consist for the most part of summary data (e.g., the number of homicides occurring in January 1998), the data from the SHR are considerably more detailed (Figure 12. Replica of Supplementary Homicide Report Form, pages 1 and 2). In some ways, the SHR can be considered a precursor to NIBRS, since it provides details about the incident's occurrence (the jurisdiction, month, and year of occurrence); the apparent circumstances under which it occurred (number of victims and offenders, whether it resulted from a robbery, domestic violence, argument, or other circumstance); age, race, and sex information about the victims and offenders, if known; and the relationship between victims and offenders, if known. As with any new data series, SHR data for the first few years had some limitations, but the series appears to be fairly accurate in terms of the number of homicides reported. The basis for this assessment is the fact that another data series on homicide exists, mortality data based on death certificates and collected by health departments since around the turn of the century. Data on Vital Statistics are collected and maintained by the National Center for Health Statistics (NCHS), US Department of Health and Human Services. Although the SHR was apparently undercounting homicides by about 14 percent during the 1960's and 1970's (Riedel, 1990), more recently the two series have been within a few percentage points of each other (Riedel, 1999). Uses of the SHR The SHR has been useful in developing policy recommendations related to homicide. Its nationwide collection, and the fact that not just the number of homicides but the characteristics of the victims and offenders are included, permits researchers to uncover patterns of significant importance: for example, that the decreasing homicide rates for some groups tended to mask the increase in homicide rates for 14- to 17-year-old males (Fox, 1997) and that infanticide is a significant problem in the United States (Maltz, 1998). In addition, many of the articles in Homicide Studies, a journal published by the Homicide Research Working Group, are based on the SHR data, and many have policy consequences. These studies may not lead to solving particular homicides, which has long been the primary focus of police attention to homicide; however, insofar as they point out patterns and risk factors, they contribute greatly to the public safety, the primary mission of the police. The accessibility of the SHR data is increased by the inclusion of files that contain SPSS and SAS data definition and programming statements, so that the SHR can be analyzed using these two statistical analysis packages or others that can read these formats. The SHR has been used not only to study homicide patterns but to study patterns of violent crime in general. The rationale for using homicide rates as a proxy for violent crime rates is because they are highly correlated (e.g., Blumstein, 1974); however, since there is very little unreported homicide in comparison to other crimes, using the homicide rate all but eliminates the problem of unreported crime. But this is misleading, and the SHR has been misinterpreted by researchers and journalists in their search for patterns in homicide data. Homicide is not so much a crime in itself as it is the fatal outcome of different crimes or "homicide syndromes," and analysis of homicide as a single entity can produce misleading results. The easy accessibility of the SHR data, then, has unfavorable as well as beneficial consequences. A better way to look for patterns in homicide data is to consider the various circumstances under which homicides occur, that is, to disaggregate infanticides from felony homicides from spousal murders, and to consider the homicide rate from within the context of the underlying crime. From this type of analysis one can investigate the risk of death due to child abuse, armed robbery, or domestic violence (Maltz, 1976, 1998; Maxfield, 1989; Block and Block, 1992). This, then is one of the great benefits of the SHR: because it provides detailed information about each homicide, it can be used to great advantage in exploring offense patterns and public policies. For example, if the risk of death due to child abuse is much higher in one jurisdiction than another, it may be that the true rates are the same but that the lower-rate jurisdiction has better child abuse reporting practices. Incomplete Provision of SHR Data by Police Departments The SHR has a number of shortcomings, in particular with respect to incomplete data. There are three ways in which SHR data may be incomplete. First, not all homicides reported on the UCR are reported on the SHR form. Second, some agencies do not include all the information about offender characteristics or motivations that is available to them. Third, even when the information is complete it may be wrong, because offender-victim relationship is given instead of victim-offender relationship or because the same relationship is given for all victims and/or offenders in an incident with more than one victim and/or offender. The FBI tries to ensure that all homicides reported on the UCR are reported in the SHR as well, by specifically requesting this information from jurisdictions for each UCR-reported homicide. The FBI doesn't always obtain it, but the SHR/UCR ratio has run between 86 percent and 96 percent between 1980 and 1994 (Snyder, 1996: 10-11). Incomplete reporting of SHR data is a greater problem. For example, the data element "Circumstance" reflects the nature of the homicide as far as it can be determined. See table 3. Yet, the number of homicides with unknown circumstances varies considerably from agency to agency, indicating that departmental policy more than knowledge of the circumstances governs the information collected by the SHR. ---------------------------------------------------------------- Table 3. Circumstances under Which Homicide Occurred Circumstances coded in SHR records 2 Rape 32 Abortion 60 Other 3 Robbery 40 Lover's triangle 50 Victim shot in hunting accident 5 Burglary 41 Child killed by babysitter 51 Gun-cleaning death not self-inflicted 6 Larceny 42 Brawl due to influence of alcohol 7 Motor vehicle theft 43 Brawl due to influence of narcotics 52 Children playing with guns 9 Arson 44 Argument over money or property 53 Other negligent handling of gun 10 Prostitution and commercialized vice 45 Other arguments 59 All other negligent manslaughter except traffic death 17 Other sex offense 46 Gangland killing 70 Suspected felony-type 18 Narcotics and drug laws 47 Juvenile gang killing 80 Felon killed by police 19 Gambling 48 Institutional killing 81 Felon killed by private citizen 26 Other felony-type not specified 49 Sniper attack Source: Fox, 1997 ---------------------------------------------------------------- There may be a number of reasons for agencies not providing complete information. First, the information may not be readily available initially, when the officer first completes the agency's homicide report -- and it may be this initial form that is used to complete the SHR form. There are strong indications that the Washington, DC, Metropolitan Police Department may fill the form out based on this preliminary report -- for example, in 1994 the offenders in 96 percent of homicides were listed as of unknown age (this was used as a proxy for "offender unknown"). Although this is the most extreme example of inadequate data collection efforts, figure 13 (based on data from Snyder, 1996) shows that Washington is far from alone. (Figure 13. Percent of Offenders with Known Ages (as Proxy for Known Offenders) in SHR Data in Counties with More than 100 Homicides in 1994) Second, some departments may downplay the utility of such information and give it low priority, since it is a voluntary collection system. Thus, the goal of obtaining complete information for crime prevention purposes too often takes a back seat to reducing the paperwork burden for a police department. Third, one city (Boston) does not provide information about the offender or his/her possible motivation "in order to prevent creating documentation that would be discoverable and of potential use to the defense at trial" (Braga, Piehl and Kennedy, 1997). (Footnote: I suspect that this policy was instituted after the information was used successfully in acquitting a defendant; I also suspect that a less drastic step could have been taken. In any event, this problem may be mooted by NIBRS, which allows a window of 2 years in which to update an incident with additional information.) Fourth, there is a great deal of variability from city to city in the diligence with which the SHR information is provided. During the 1997 workshop it was mentioned that one city (Washington, DC) rarely records drug involvement in homicides, while in another city (Detroit) almost every homicide is recorded as drug-involved -- when the actual truth for both cities is somewhere in between. Moreover, incompleteness in SHR reporting also reflects the coding procedures established by the FBI to collect the data. The codebook (Fox, 1996) for the SHR data gives an example: "[T]he structure of the data collection forms prescribes that the relationship of the offender to the first victim (often chosen arbitrarily) be coded for this offender. Thus, for example, in 1977 a Redondo Beach, California, woman killed her husband and three step-children by burning down the family home. Appropriately in this case, the weapon was coded a 'fire' for all four victims, but the relationship of victim to offender was coded as 'step-daughter' for all victims -- two 8-year-old white females, a 7-year-old white male, and a 40-year-old white male." That is, the FBI strips the relationship data that may be provided to the FBI and uses only one relationship to characterize the entire incident. This problem does not affect most homicides, however, since the great majority of homicides consist of one victim and one offender. Updating SHR Files SHR files are updated when additional information is provided by police departments. It should be understood that the SHR file is updated, but individual records are not updated. For example, an accidental death may be reclassified as a homicide, and consequently is sent in to the FBI for inclusion in the SHR file; it is in this way that the SHR file is updated. However, if police submit an SHR record to the FBI with a homicide whose offender was classified as unknown, and subsequently learn of the identity of the offender (the Unabomber or Theodore Kasczinski case is exemplary), the records of those homicides are not changed. The FBI cannot revisit old records because no unique index code is included in the SHR file that would permit them to identify specific homicides. This problem will diminish when NIBRS is implemented, since each incident will have a unique identifier, permitting true updating for 2 years after the incident. Insofar as police departments adopt NIBRS and adhere to its requirements, it will be possible to truly update the incident files as more information about incidents develops. Availability of SHR Data Sets The FBI makes the data files available to BJS, and the files are then restructured, reformatted, cleaned, and given wider accessibility through the NACJD website (at http://icpsr.umich.edu/nacjd/ucr.html). Each record can contain information on up to 11 victims and offenders; since most homicides are one-victim/one-offender homicides, this makes each year's file much larger than it need be and consequently more difficult to analyze. In addition, there is a separate file for each year, so these datasets need to be combined to perform any multi-year analyses. For this reason, BJS and the National Institute of Justice funded an effort to make the SHR data more accessible, resulting in a multi-year SHR data set, 1976-94 (Fox, 1996). Included in the restructured file are weights that take into account missing offender data according to their age, race, and sex, at both the State and Federal levels. Manslaughters by negligence and justifiable homicides are not included in the data set. There are problems in dealing with multiple victims and offenders in a single data set, in a way that keeps the victim-offender relationships intact without having to carry along (mostly empty) space for 11 victims and offenders. The way it is handled in the 1976-94 dataset is to create a different record for each victim-offender pair. That is, an incident with four victims and two offenders would have eight records, each record corresponding to a different victim-offender pair. Multi-year SHR data sets from 1976-97 will soon be available at NACJD, which handle multiple victims and offenders in a different manner. Two separate data sets are generated from the FBI data files, a victim data set and an offender data set. The victim data set contains a separate record for each victim; if a single homicide incident includes four victims and two offenders, four records are created -- one for each victim -- and the offender data included on those four records are the characteristics of the first offender. To describe the same incident, the offender data set would include two records, one for each offender, and the victim data included on those two records are the characteristics of the first victim. SHR Imputation The first point to be made about imputation of the SHR is that the FBI does not impute SHR data. However, the victim and offender data sets for the combined 1976-97 SHR data, to be provided at NACJD, do include imputation procedures. The imputation procedures incorporated in these data sets are not the only ones that have been used for SHR data, but because they are used on the most complete SHR data sets (1976-97) -- and the ones most likely to be used in the future -- their characteristics are described below. Two different kinds of imputation are used in the (to-be) archived multi-year SHR data set. The first one is used to reconcile the count of SHR homicide victims with the count in CIUS. The second imputation procedure is used to estimate the characteristics of offenders in incidents in which there is no information about the offender. In both types of imputation weights are assigned to each case. The best way to explain these imputation procedures and their use is to discuss each type of weight given in the SHR file. Weighting the Victim File The number of records in the victim file is the count of SHR homicides. As noted earlier, this number is often not the same as the count of UCR homicides, both nationally and at the State level. Two of the weights included in the victim file are used to reconcile these two numbers. Weight wtus. This weight is the same for all cases for a given year. The weight represents the ratio of the number of homicides reported in CIUS to the number reported in the SHR. Thus, since the UCR reported 18,780 homicides and the SHR reported 16,605 homicides for 1976, the weighting factor wtus is 1.13 (18,870/16,605) for 1976. It is used in the following way: suppose one wants to estimate the number of homicide victims under 6 years of age. The UCR does not detail this information, but we can estimate the number by extrapolating from the known SHR cases to the UCR cases. In 1976 the SHR recorded 519 such cases, so the 1976 estimate would be (1.13 x 519 =) 587 victims under age 6. This would permit us to compare 1976 data with data from another year, in which a different weighting factor is used. For example, in 1977, 548 such cases were recorded in the SHR, representing a 6% increase over the 519 in 1976. However, wtus for 1977 was 1.06 (19,120 UCR homicides versus 18,032 SHR homicides), so we estimate that there actually were 581 such victims; this represents a 1% decrease from the 1976 estimate of 586. (Footnote: These examples are given for illustrative purposes only. A better way of estimating year-to-year change is to obtain the rate, by dividing the estimates by the estimated population under 6 in each year.) ------------------------------------------------------------- Table 4. Victim Imputation in the SHR Estimated UCR SHR UCR/ SHR age UCR age Year total total SHR ratio 5 or under 5 or under 1976 18780 16605 1.13 519 587 1977 19120 18032 1.06 548 581 Victims under age 5 -- SHR percent increase 5.6% Estimated UCR percent increase -1.0% ------------------------------------------------------------- Weight wtst. This weight is the same for all cases in a State. The weight represents the ratio of the number of homicides reported by the State in CIUS to the number reported In the SHR. Thus, the weighting factor wtst of 1.17 for Alabama's 453 SHR-recorded homicides for 1976 indicates that Alabama experienced (1.17 x 453 = ) 530 homicides in 1976. For Alaska, however, wtst was 1.0 for all of the 43 homicides, indicating that the UCR and SHR reported the same number of homicides. In 1977 the values of wtst for Alabama and Alaska were 1.06 and 0.96, respectively, so the SHR counts of 487 and 46 indicate that these States had (1.06 x 487 =) 516 and (0.96 x 46 =) 44 UCR-reported homicides, respectively. (Footnote: There may be more SHR homicides than UCR classifications due to crime reclassification across years. For example, perhaps two shooting victims in Alaska in incidents classified as aggravated assaults in 1976 died in 1977.) Weighting the Offender File There are three weights in the offender file. Just as the weights in the victim file are meant to provide a better estimate of the number of homicide victims, the weights in the offender file are meant to provide a better estimate of the number of homicide offenders. Whereas the victim file weights are used to fill in for missing records, the offender file weights are used to fill in for missing data within records, that is, for cases where the identity and characteristics of the offender are unknown. Weight wtimp. This weight imputes, for the Nation at large, the number of offenders by age/race/sex category. Suppose that there are 500 victims in a specific age/sex/race category, and 400 of them are killed by known offenders. Then wtimp would equal 1.25 because the unknown offenders are presumed to have the same age/sex/race characteristics as the known offenders. For example, suppose that 25 percent of them (or 100) are killed by white males ages 15-24. Then we would estimate that 25 of the victims of unknown offenders are also killed by white males ages 15-24. Weights wtimpus and wtimpst. In order to take into account the cases that are reported to the UCR but not included in the SHR, we can estimate the total number of white male offenders ages 15-24 in the United States, by multiplying wtimp by the aforementioned wtus. If the latter is 1.13, as in the earlier example, the value of the weight wtimpus would be (1.13 * 1.25 =) 1.41. Similarly, if we wanted to estimate the number of such offenders in Alabama, we would weight each homicide committed by a 15- to 24-year-old white male by (1.17*1.25 =) 1.46, the value of wtimpst for Alabama. Problems with this SHR Imputation Procedure Since there are two different weighting schemes, for the victim and offender data sets respectively, the inaccuracies that arise from their application need to be considered separately. Victim File. The weights wtus and wtst are applied to reconcile the number of homicides recorded in the UCR and SHR, at the national or State levels, respectively. When there is a discrepancy between the two, the UCR is usually (but not always) the larger of the two, so these weights are usually larger than 1. There may be a number of reasons for the discrepancy. *There may be clerical errors in an agency's submission of either Return A or the SHR form. For example, some homicides that were reported to the UCR may have "slipped through the cracks" when an agency filled out the SHR form, or perhaps because the agency did not fill the form out at all. If the former is true, then the agency is likely to be a larger police department with many cases to report; if the latter is true, then the agency is likely to be a rural department that does not report homicides very often. *Part of the discrepancy may be due to date slippage, as when a person is injured in one calendar year and succumbs to these injuries in another year. In such cases, the homicide may not be reported on the SHR form, especially if agency practice is to fill it out at the time of the offense. In any case, assuming that the missing homicides generally have the same characteristics as the reported ones (which this weighting scheme implies) may be in error. The error attributable to this, however, is likely to be small, since the concordance between the two data sets is usually fairly high (Chilton and Jarvis, 1999). Offender File. The weights wtimp, wtimpus, and wtimpst estimate the number of offenders in cases where the offenders are unknown. However, there is a problem in assuming that the best means of estimating offender characteristics is to predicate the estimation on the age, race, and sex of the victims. For example, suppose that most 45-year-old women are killed by their mates, i.e., by 40- to 49-year-old white males. However, if a 45-year-old female clerk at a convenience store is killed by an unknown assailant, not by her husband, there should be some way of including this in the imputation equation. In other words, the circumstance of the killing (which is included in the SHR record) is probably a better indicator of offender characteristics than a rule that does not include this information. Williams and Flewelling (1987) used this imputation method; however, Langford, Isaac, and Kabat (1998) describe some limitations to using circumstance for imputation purposes. An indication of the problem with this type of imputation is made clear when investigating the distribution of unknown homicides within a State. Between 1976 and 1996 Richmond accounted for 18 percent of Virginia's homicides but 41 percent of its homicides by unknown assailants; Atlanta accounted for 29 percent of Georgia's homicides but 49 percent of its unknowns; and Indianapolis accounted for 21 percent of Indiana's homicides and 35 percent of its unknowns. (Footnote: These figures are based on an analysis by the author.) So making the assumption that the unknown assailants have the same characteristics as the known offenders is questionable. More to the point, the very fact that the assailant is unknown often means that the type of homicide is quite different than those in which the assailant is known. Studies in Chicago and Boston confirm this assertion. Block (1998) compared 1993-94 Chicago homicides found in the SHR data with those in the Chicago Homicide Dataset (CHD), perhaps the most complete homicide dataset in the country. Her data show that about 10 percent of the CHD incidents had unknown offenders in the SHR data, but that it varied from around 30 percent (for sexual assault) to under 4 percent (when firearms were used). Braga et al. (1997) compared Boston SHR data for 1990-94, for victims 21 or under (N = 155), with data they had collected for this population in conjunction with the Boston Gun Project (BGP -- see Kennedy, Piehl, and Braga, 1996), a program to reduce youth gun violence. Since the BGP data were collected well after the events, they were much more detailed than the SHR data, especially considering the concerns of the Boston homicide detectives (see page 34 above). Thus, it was not surprising to find that, while the SHR data listed 65 percent as with unknown circumstances (and 79 percent with unknown victim-offender relationship), the BGP data had only 30 percent -- less than half as many -- with unknown circumstances and 40 percent -- again about half as many -- with unknown victim-offender relationship. Although neither city is representative of the Nation as a whole, what these studies point out is that the unknown offenders are not necessarily representative of the knowns. This may be due to the fact that homicide is not a specific crime like robbery. Rather, as mentioned earlier, it is the fatal outcome of a lot of different crimes: intimate partner violence, armed robbery, child abuse, etc. The imputation rule used in the SHR offender file implicitly assumes that they are all of the same general nature, which is not the case; moreover, the information included in the SHR file is specifically included so that different types of homicides can be distinguished from each other, and an imputation rule for the SHR should take this information into account. Suggested Alternative SHR Imputation Procedure A homicide's circumstance (see table 3), rather than the victim's characteristics, would seem to be a more appropriate indicator of the characteristics of the offender. No matter what the age, race, and sex of the victim, if the homicide arose during the course of a robbery the characteristics of the offender would probably resemble other robbery- homicide offenders more than they would any other class of offender. To some extent, this suggestion takes the SHR reporting practices of some agencies into account: even though Washington, DC, reported that 94 percent of the 473 homicides had unknown offenders, the circumstances were listed as unknown in only 50 percent of the cases. However, this suggestion should be considered carefully. Maxfield (1989) noted the lack of consistency among agencies in coding homicide circumstance. Because of its importance, imputation of homicide offenders is a topic that should not be undertaken lightly. In the aggregate it may not amount to much, but as we begin to go beyond looking only at aggregate rates and investigating small subgroups it takes on more importance. Newspapers report not just on State-by-State homicide rates, but on offense rates, by State, by age group, by weapon, and the rates inferred by imputation may be seriously in error, as suggested by the Boston and Chicago studies. Granted, the "unknown" offenders in Boston and Chicago do not represent unknown offenders in the rest of the United States; these cities were chosen because of the availability of good data; in both cities homicide detectives and researchers have formed a strong working relationship. Rather, as noted earlier, despite their lack of representativeness, these studies do provide a benchmark against which to test different imputation procedures. Only through research like this can we find the extent to which our view of crime is distorted by incomplete and inaccurate data. X. Conclusions and Recommendations Reporting Practices The primary focus of this report has been to understand, and make suggestions for improving, how the FBI collects and analyzes crime and arrest data that are subsequently published in CIUS. Along the way, however, one could not help but notice the recent decline in the quality and quantity of the data submitted by State and local agencies to the FBI. This trend compromises the quality of information we have about the nature and extent of crime in the United States. I hope that this report serves as an impetus to improve the completeness and accuracy of crime and arrest data. One State, Illinois, sends data to the UCR program that do not adhere to UCR reporting standards with regard to the hierarchy rule. This occurred for the best of reasons: in 1992 Illinois embarked on an ambitious effort to implement NIBRS statewide, but the complexities were too great, and the effort was suspended in 1994. Beginning with 1993 data and continuing to the present, Illinois has submitted only summary UCR data, but without applying the hierarchy rule, so that incidents with multiple offenses may be counted more than once. In addition, since 1984 Illinois statutes have defined forcible sexual assault without reference to gender. This is not compatible with UCR standards. Because of this incompatibility, the UCR does not include Illinois data on rape. Illinois, one of the most populous States (and my home State), should be encouraged to change the reporting practices that have kept it from contributing to the UCR in the recent past. Although this issue was not addressed in the report, and is somewhat beyond its scope, the absence of crime and arrest data from Federal agencies distorts the picture we have of crime in the United States. The recent report detailing the high rate of victimization of American Indians (Greenfeld and Smith, 1999) serves to underscore this deficiency in our crime statistics: insofar as UCR data are used to allocate resources, this would affect the extent to which enforcement resources for American Indians are provided. Publishing and Archiving It is a tribute to the diligence and dedication of the FBI's Program Support Section staff that there are so few errors in a report so filled with numerical data as CIUS; one cannot just run a spell-checking program to see if errors have been made in the manual transcription of numbers from one medium to another. However, it is a waste of personnel time to use a process that requires so much staff time to proofread numerical data. The tables were initially produced by a computer that could, with relatively little additional funding and effort (and that only for the first year) be set up to produce camera-ready output. This added technology would not only remove an unneeded burden from the PSS staff, but it has the promise of reducing the time taken to produce CIUS by a matter of some weeks, perhaps even months. This extra time could be put to use at either end of the report production process: ORIs and States could be given more time to transmit their data to the FBI, which may become increasingly important as more agencies convert to NIBRS; and/or the report could be published earlier in the year. It should be noted that two other annual reports, Hate Crime and Law Enforcement Officers Killed and Assaulted, are published directly by the PSS staff; the FBI should consider direct publication for CIUS as well. If it is possible for the FBI to do so, BJS should request the FBI to "freeze" the version of the raw data set that is used to produce CIUS, and send it to NACJD for archiving, as well as the final version of that data file. This will permit researchers to perform data analyses that are consistent with CIUS and/or use the most current data to provide more complete analyses. Imputation As this report details, imputation of crime and arrest data has been based on ad hoc procedures that were appropriate at the time they were made and for the uses to which they were originally put. Now that UCR data are being used for different purposes, new methods of imputing data need to be considered. This report describes some reasonable candidates, but these should not be applied without setting up an evaluation program to determine whether they actually provide improved estimates. This is true for all data sets in which imputation is being used: crime, arrest, and SHR data. One area where some experimentation might be helpful is to determine where the cutoff threshold should be before an ORI's data are sufficient to be included without full imputation, the 3 months used by FBI or the 6 months previously used by NACJD. Among the possible ways of performing these experiments are: *Analyze nonreporting agencies' submissions to their municipal governments to see to what extent the imputation procedure is valid. *Develop different categories of nonreporting ORIs and randomly select a number of full reporters from each category; apply an imputation method and see how closely the imputed data comes to the actual data. *Do the same with incomplete-reporting ORIs. The data from so-called "zero-population" agencies should also be imputed. The exact means of doing so should probably depend in part on the nature of the jurisdiction: the procedure for imputing a State police agency's data should not be the same as for a university police department. NIBRS NIBRS represents a major increase in the amount of data to be collected by local agencies and forwarded to the FBI. A number of States have implemented NIBRS successfully in essentially all their reporting agencies; however, others have had software and other problems that have not only prevented NIBRS from being fully implemented, but they have been unable to send even the summary UCR data to the FBI. Some police administrators have complained about NIBRS and the level of detail it entails. However, its implementation is in part a recognition that if the focus of policing is to be more than just "catching the bad guys," and to deal with public safety more generally, then the police need to be concerned with analyzing crime for patterns that go beyond the modus operandi of individual offenders. Such a pattern may suggest, for example, that a new domestic violence or after-school program might reduce certain types of offenses. Implementing NIBRS will give us the ability to compare the effectiveness of such programs in different jurisdictions with different populations. A recent FBI report, The Structure of Family Violence (found on the FBI's website at www.fbi.gov/ucr.htm/famvio21.pdf), gives some indication of the way NIBRS data can be used for this purpose. As I mentioned in the beginning of this report, my goal (and the goal of those who attended the workshop) has been to suggest some ideas for consideration in revising the FBI's Uniform Crime Reporting Program. While not all of the suggestions may be feasible to implement at this time, I hope that this report lays the groundwork for improving our knowledge of the nature and extent of crime in the United States, one of the more pressing social problems confronting the Nation. Appendix A: Persons Attending the Workshop on UCR Imputation Procedures US Department of Justice Participants: Bureau of Justice Assistance Richard Ward Bureau of Justice Statistics Jan Chaiken Lawrence Greenfeld Charles Kindermann Patrick Langan Sue Lindgren Michael Rand Bruce Taylor Marianne Zawitz Criminal Division Julie Samuels Steven Shandy Federal Bureau of Investigation Yoshio Akiyama Ben Brewer Gilford Gee Antonio Hwang John Jarvis Dawn Kording Victoria Major James Nolan Sharon Propheter National Institute of Justice Jordan Leiter Office of Juvenile Justice and Delinquency Prevention Barbara Allen-Hagen Other Participants: American University James Lynch Inter-University Consortium for Political and Social Research Nora Arato Christopher Dunn Christopher Lysholm Kaye Marz Massachusetts State Police Daniel Bibel National Center for Juvenile Justice Howard Snyder Northeastern University James Fox Research Triangle Institute Robert Flewelling Rutgers University Michael Maxfield Project SEARCH David Roberts University of Illinois at Chicago Michael Maltz University of Massachusetts, Amherst Roland Chilton Appendix B: State On-Line Publication of Crime Data Data reported on website Law enforcement Crime Officers killed/ State UCR Crime trends Hate/bias Arrests assaulted Personnel Website URL Alabama 97 93-97 x x x http://agencies.state.al.us/ acjis/pages/alacrime.htm Arkansas 95-96 97-98 x x x http://www.acic.org/ statistics.htm California 96-97 x x x x http://caag.state.ca.us/cjsc/ pubsol.htm Colorado 80-96 http://www.state.co.us/gov/ dir/cdps/dci/ors/stats.htm Connecticut 96-97 http://www.state.ct.us/dps/ CT-UCR.htm Florida 97 96-97 x http://www.fdle.state.fl.us/ Crime_Statistics Georgia 98 85, 95-96 x http://www.ganet.org/gbi/ stcrime.html Hawaii 83-97 92-97 x x x http://www.cpja.ag.state.hi.us/rs/ Illinois 94-97 93-97 x x x x http://www.state.il.us/isp/ cii00001.htm Iowa 96 x x x http://www.state.ia.us/ government/dps/crime/stats/ Kentucky 95-96 x x x http://www.state.ky.us/ agencies/ksp/crime.htm Massachusetts 95 http://www.magnet.state.ma.us/ msp/crimedat.htm#Taunton Michigan 97 88-97 x x x x http://www.state.mi.us/msp/ crd/ucr/contents.htm Minnesota 95-96 http://www.dps.state.mn.us/bca Nebraska 93-95 http://www.info.ded.state.ne.us/ stathand/contents.htm New Hampshire 97 http://www.state.nh.us/nhsp/ New Jersey 93-97 88-97 x x x http://www.state.nj.us/lps/ njsp/stats.html New York 90-96 x http://criminaljustice.state. ny.us/crimnet/pubs.htm North Carolina 93-98 78-97 x x x http://sbi.jus.state.nc.us/crimstat/ nccrime.htm Ohio 80-96 80-96 http://www.ocjs.state.oh.us/ Pennsylvania 76-96 x http://www.state.pa.us/PA_Exec/PCCD/stats/factsheets/statspag Utah 96 x x x http://www.ps.ex.state.ut.us/ Vermont 97 x x http://www.dps.state.vt.us/ cjs/crimestats.htm Virginia 97 96-97 x x x http://www.state.va.us/vsp/ zucr1.html Appendix D: Extent of UCR Data Coverage, Alabama-Wyoming, 1958-97 (Data for these 50 graphs are available in a spreadsheet.) Appendix E: Missing Data in UCR Files Used for the 1996 LLEBG Formula Calculations Sue Lindgren Bureau of Justice Statistics January 23, 1997: Modified for the April 24-25, 1997 Workshop on UCR Imputation Procedures Background The Local Law Enforcement Block Grant (LLEBG) program allocates Federal funds to State areas and local governments based on a 3-year average of Part I violent crimes as reported to the FBI. The Bureau of Justice Assistance (BJA) administers the program, using award amounts computed by the Bureau of Justice Statistics (BJS). The legislation recognizes that some jurisdictions do not report to the FBI and prescribes that we estimate data for such "nonreporting agencies." Because of the press of time and our unfamiliarity with the Uniform Crime Reporting (UCR) data base to be used in the formula, it was not possible to attempt any such estimates for the 1996 awards (computed late spring, early summer 1996 after the formula was enacted in April 1996). Rather, we assumed that there would not be a significant number of such "nonreporting" agencies eligible for an award and decided that we would handle such situations on a case-by-case basis. A few such cases were brought to our attention and we researched and resolved each on an individual basis. Once the formula was run and the grants awarded, and State and local government inquiries about award amounts researched and resolved, we began the following analysis to inform BJS and BJA of the extent and nature of missing data in the UCR data set. Extent of Missing Data, by State As the first part of this analysis, we compared the violent crime State-wide totals we computed by summing all agency records within each State on the UCR files to the estimated violent crime totals published by the FBI in table 5 of Crime in the United States for the 1992-94 period used in the 1996 formula awards and for the 1995 data. (Footnote: Earlier years data were used in the fiscal 1996 formula data set for Kansas, Illinois, and Montana because of their conversion to NIBRS. For those States, the fiscal 1996 analysis examined data availability for those earlier years.) As seen in the attached tables 1-3, there is considerable variability across States. For the 1992-94 period, three States had less than 70% complete data, and 11 had less than 90%. By 1995, 28 States had even lower coverage than in the 1992-94 period, although the 1995 data for at least 4 States are low because of conversion to NIBRS. (We used data prior to 1994 for three of these States in the 1996 awards). Characteristics of Agencies with Less Than 36 Months of Violent Crime Data The second part of this analysis examined what kinds of agencies are included in the UCR files, but, according to the FBI, reported zero months for the period used for the formula. The FBI files contain a variable giving the number of months reporting for each year; using this variable for the three years, we were able to construct a "Number of months reporting" variable with a range of 0 to 36 months. Of 18,413 agencies in the FBI's UCR files for 1992-94, 3,516 (19.1%) did not report for any month during the 36-month period used in the formula and another 3,197 (17.4%) reported between 1 and 35 months. (See attached table 4.) Agencies reporting 0 months and agencies reporting 1 to 35 months are examined separately below. Agencies with 0 Months of Data We looked first at the 3,516 agencies reporting for no months during the 36 month period used in the formula. Based on our analysis, we discovered that many of the agencies on the files are now "retired" or "inoperative." This is less important for imputation purposes than for formula purposes, where we need to be sure we are not estimating data for agencies no longer in existence. Despite the research described below, we are not in a position to say with certainty which are inactive and which are nonreporting, although the research gives some idea of the relative sizes of each group. As seen in table 5, almost half, 1,652, of those reporting no months were covered by another reporting agency by the end of 1994, making them ineligible for an LLEBG grant by our interpretation of the legislation, and a third, or 1,221, are "zero population" agencies, meaning they are almost certainly State agencies or special police agencies (such as transit police) that probably are not eligible for a formula award. (194 of those were covered by another reporting agency as well as being zero population.) Excluding all of these presumably ineligible agencies, we are left with 866 agencies coded as reporting no months, most of which are small jurisdictions. As seen in table 6, nearly 70% are cities with less than 10,000 population. However, when looking at the actual jurisdictions, there are 2 counties and 1 city with over 100,000 population, 3 cities with between 50,000 and 100,000, and 199 cities and counties with over 10,000 population. For formula allocation purposes, it is important to keep in mind that in less populous States, cities with 10,000 population receive awards; thus, there is more of a potential impact here than we originally thought. Jurisdictions with between 1 and 35 Months of Data Table 4 shows that 17.4% of the agencies in the UCR files reported for at least 1 month, but less than 36. This represents 3,197 agencies. As seen in table 7, 17% of these, or 554, are zero population agencies, again meaning they are probably not eligible jurisdictions, including 5 that were covered by another reporting agency by the end of 1994. An additional 50 with populations are covered by another reporting agency, again indicating ineligibility. Of the remaining 2,593 that might be eligible for an LLEBG grant, 10%, or 262 jurisdictions, already exceeded their State threshold for an award; their awards would increase if they reported for more of the period. (Footnote: This includes three State police barracks that were not "zeropop;" that is, unlike most State-level agencies they had an entry other than zero in the "population covered" variable. See "Addendum on zeropop agencies." ) Most of the 2,331 that did not get an award would not reach the threshold if they reported fully; 525 of them reported no violent crime, so it is unlikely that they would have enough violent crime in the months they did not report to get an award. (Footnote: Zero violent crime is a possibility; the agency may have reported property crime (which is not on the files supplied to us), or may have no crime at all but continues to participate in the UCR program. Interestingly, 792 fully-reporting agencies report no violent crime; 484 are zero population, and 308 have populations.) Of those reporting some violent crime, the overwhelming majority are so far below the threshold for an award in their States that it seems unlikely that they would receive an award even if they had complete data. However, 28 of the 2,331 had over 90% of the threshold amount, and another 26 had between 80% and 90%, and more complete data may have resulted in an award. Impact of Incomplete Data It is clear that a considerable number of agencies are adversely affected in the formula by missing data in the UCR data sets; data that are missing for a variety of reasons beyond not participating in the UCR program. In the course of responding to State and local inquires, we heard some anecdotal evidence from individual police departments who claim they submitted data to the State program that are not showing up on the UCR files. For formula purposes, the reason for missing data is a concern, but not necessarily for imputation purposes. While estimating the total impact of missing data on the formula allocations is not attempted, a few facts are clear: * 262 jurisdictions that received an award would get a larger award with more complete data coverage. * It is possible to estimate that, if all cities reported at the same rate as those that did for 36 months, an additional 84 cities would receive an award. * Because of the configuration of the award files, it is not possible to replicate that computation for counties, but presumably some of the 2,212 counties that did not get an award would with more complete data coverage. Although the impact of incomplete data seems greatest for larger jurisdictions which may have hundreds of violent crimes a year, each of which brings $216.62 in most States, small jurisdictions in those States with very low thresholds such as Vermont (4 violent crimes) and North Dakota (7 violent crimes) might quickly become eligible for an award with more complete data. (Footnote:The LLEBG legislation establishes $10,000 as the minimum local award. Thus, a jurisdiction needs a certain amount of violent crime to qualify for an award. In some States this amount is lower than the national average because the State-level formula establishes a minimum State award regardless of violent crime levels.) In addition, the dollars per crime figures for those States are considerably higher: $2,321.17 per crime in Vermont and $1,509.17 in North Dakota. In reviewing these results, it is necessary to keep in mind that because the formula is relative and interdependent, an increase in reporting (or in estimating missing data) would result in an increase in the thresholds and a decrease in the dollar value of a single crime. Addendum on "Zeropop" Agencies Agencies without an entry in the "population covered" variable on the UCR tapes were coded to be "zeropop" in the variable by that name. This was used as a surrogate indicator to identify State police, special police, and similar agencies that were not a city or county police or sheriff's office that might be eligible for a local law enforcement formula award. This was necessary because the UCR coding scheme does not allow the separation of State police and university police from county and city police. In the process of examining the extent of missing data in the UCR files, it was discovered that some of these agencies actually had a population entry and thus were not coded "zeropop." A search of the file for "State Police" in the agency name uncovered 136 of these with a population figure, about 15% of agencies with "State Police" in the name. These are primarily State police barracks in various counties reporting data for the county area. This had no effect on the distribution of fiscal 1996 awards because any agency not coded as a city was treated as reporting for a county. It will not affect the fiscal 1997 awards because they will rely on the Census Bureau government identification number that is being added to the files to allow accurate identification of city, county, and State agencies and special police forces. This should have no impact on the imputation of county-level data because we will be able to provide a crosswalk between the FBI's agency codes (ORI codes) and the Census Bureau government identification, level-of- government, and type-of-agency codes so that those doing the imputation can treat those agencies any way they want. Table 1. 1992-94 UCR Total Violent Crime Data for Input to 1996 LLEB (Numbers on data file as percent of estimated amounts presented in Crime in the U.S.) State Percent State Percent South Carolina 101.57 % Minnesota 97.74 Idaho 100.49 Colorado 97.56 Wyoming 100.15 Arizona 97.5 Connecticut 100.01 Nebraska 97.48 Rhode Island 100.01 Alaska 96.39 Hawaii 100 Georgia 95.52 New Jersey 99.99 Missouri 95.41 Maryland 99.98 Illinois 95.32 Texas 99.97 Pennsylvania 94.99 West Virginia 99.89 Nevada 94.49 California 99.87 Michigan 94.13 Oklahoma 99.86 Tennessee 90.36 Virginia 99.81 Alabama 90.24 New York 99.73 Ohio 90.07 Wisconsin 99.70 South Dakota 89.65 Arkansas 99.66 Louisiana 89.27 Oregon 99.43 New Hampshire 87.37 Maine 99.31 Massachusetts 86.33 North Dakota 99.30 Indiana 82.29 Washington 99.13 New Mexico 75.81 Kansas 98.98 Iowa 74.56 Kentucky 98.45 Vermont 74.35 Florida 98.37 Mississippi 68.04 Utah 98.33 Delaware 65.35 North Carolina 98.24 Montana 63.00 Table 2. 1992-94 Summary of Actual Counts of Violent Crime on LLEBG Data Files as a Percent of the Amount Estimated by the FBI Part I violent crimes on data Percent of estimated Cumulative files as a percent Number of number of Part I percent of published estimate States violent crimes of estimated crime 100% or more 6 12% 12% 99% to 99.99% 14 28 40 98% to 98.99% 5 10 50 96% to 97.99% 5 10 60 90% to 95.99% 9 18 78 80% to 89.99% 5 10 88 Less than 80% 6 12 100 50 100 Table 3. Comparisons of UCR Violent Crime Data for Input to 1996 and 1997 LLEBG F (Numbers on data file as percent of estimated amounts present in Crime in the U.S. Part I violent crimes as percent of published Difference in the percentage estimates, 1992-94 from 1992-94 to 1995 1995 Alabama 90.24% 95.09% 4.85% Alaska 96.39 94.18 -2.21 Arizona 97.50 96.87 -0.63 Arkansas 99.66 99.78 0.12 California 99.87 97.19 -2.68 Colorado 97.56 97.12 -0.44 Connecticut 100.01 99.92 -0.09 Delaware 65.35 35.51 -29.84 Florida 98.37 98.75 0.38 Georgia 95.52 96.39 0.87 Hawaii 100 100.00 0 Idaho 100.49 99.17 -1.32 Illinois* 95.32 63.03 -32.29 Indiana 82.29 75.05 -7.24 Iowa 74.56 79.03 4.47 Kansas* 98.98 34.21 -64.77 Kentucky 98.45 91.51 -6.94 Louisiana 89.27 94.50 5.23 Maine 99.31 99.63 0.32 Maryland 99.98 99.99 0.01 Massachusetts 86.33 91.49 5.16 Michigan 94.13 94.96 0.83 Minnesota 97.74 99.47 1.73 Mississippi 68.04 61.41 -6.63 Missouri 95.41 94.74 -0.67 Montana 63.00 17.92 -45.08 Nebraska 97.48 99.07 1.59 Nevada 94.49 99.32 4.83 New Hampshire 87.37 75.19 -12.18 New Jersey 99.99 100.09 0.10 New Mexico 75.81 73.24 -2.57 New York 99.73 98.70 -1.03 North Carolina 98.24 98.83 0.59 North Dakota 99.3 95.32 -3.98 Ohio 90.07 83.98 -6.09 Oklahoma 99.86 99.84 -0.02 Oregon 99.43 97.90 -1.53 Pennsylvania 94.99 83.95 -11.04 Rhode Island 100.01 100.11 0.10 South Carolina 101.57 102.38 0.81 South Dakota 89.65 82.35 -7.30 Tennessee 90.36 87.66 -2.70 Texas 99.97 99.78 -0.19 Utah 98.33 95.84 -2.49 Vermont 74.35 90.61 16.26 Virginia 99.81 99.88 0.07 Washington 99.13 95.90 -3.23 West Virginia 99.89 99.90 0.01 Wisconsin 99.7 99.83 0.13 Wyoming 100.15 99.34 -0.81 Maximum 1.0157 1.0238 Minimum 0.63 0.1792 Table 4. Number of Months/Years Reporting, All Agencies in UCR Files, Number of Percent of reporting agencies all agencies No months 3516 19.1% 1 - 11 months 404 2.2 1 full year 508 2.8 13 - 23 months 464 2.5 2 full years 530 2.9 25 - 35 months 1291 7.0 3 full years 11700 63.5 Total 18413 100 Table 5. Agencies Reporting Zero Months, by Whether Covered by Another Reporting Agency or "Zeropop" Covered by another agency No Yes Total Zeropop 1027 194 1221 Non-zeropop 836 1458 2294 Total 1863 1652 3515 Table 6. Population Distribution of Agencies Reporting and not Covered by Another Reporting Agency or "Zeropop" Government type and size Frequency Percent City 100,000-250,000 1 0.1 City 50,000-100,000 3 0.3 City 25,000-50,000 12 1.4 City 10,000-25,000 38 4.4 City 2,500-10,000 197 22.7 City under 2,500 398 46.0 County 100,000 or more 2 0.2 County 25,000-100,000 46 5.3 County 10,000-25,000 97 11.2 County under 10,000 72 8.3 Total 866 100 Table 7. Agencies Reporting 1 to 35 Months, by Whether Covered by Another Reporting Agency by 1994 or "Zeropop" Covered by another agency No Yes Total Zeropop 549 5 554 Non-zeropop 2593 50 2643 Total 3142 55 3197 --------------------------------------------------------------- References Akiyama, Yoshio, and James Nolan (1999). "Methods for Understanding and Analyzing NIBRS Data." Journal of Quantitative Criminology, 15, 2, 225-238. Akiyama, Yoshio, and James Nolan (1999b). "UCR Data Imputation: Longitudinal vs Cross-Sectional Approaches." Internal memorandum, FBI Criminal Justice Information Services Division, August 1999. Biderman, Albert D., and James Lynch (1991). Understanding Crime Incidence Statistics : Why the UCR Diverges from the NCS. Springer-Verlag, New York, NY. Black, Dan A., and Daniel S. Nagin (1998). "Do Right-to-Carry Laws Deter Violent Crime?" Journal of Legal Studies, 27, 1, 209-220. Block, Carolyn Rebecca (1998). An Evaluation of the Completeness and Accuracy of SHR Data: Chicago, 1993 and 1994. Illinois Criminal Justice Information Authority, Chicago, IL. Block, Richard and Carolyn Rebecca Block, "Homicide Syndromes and Vulnerability: Violence in Chicago Community Areas over 25 Years." Studies on Crime & Crime Prevention 1, 1992, 61-87. Braga, Anthony A., Anne M. Piehl, and David M. Kennedy (1997). Youth Homicides in Boston: An Assessment of Supplementary Homicide Report Data. John F. Kennedy School of Government, Harvard University, Cambridge, MA. Chilton, Roland, and John Jarvis (1999). "Victims and Offenders in Two Crime Statistics Programs: A Comparison of the National Incident-Based Reporting System (NIBRS) and the National Crime Victimization Survey (NCVS)." Journal of Quantitative Criminology, 15, 2, 193-205. Ehrlich, Isaac (1975). "The Deterrent Effect of Capital Punishment: A Question of Life and Death." American Economic Review, 65, 397-417. Federal Bureau of Investigation (annually). Crime in the United States. Federal Bureau of Investigation, U.S. Department of Justice, Washington, DC. Federal Bureau of Investigation (1958). Crime in the United States. Special Issue, containing the Report of the Consultant Committee. Federal Bureau of Investigation, U.S. Department of Justice, Washington, DC. Fox, James Alan (1996). Uniform Crime Reports: Supplementary Homicide Reports, 1976-1994. [Computer file and codebook]. ICPSR version. Northeastern University, College of Criminal Justice, Boston, MA [producer]. Inter-university Consortium for Political and Social Research [distributor], Ann Arbor, MI. Fox, James Alan (1996). Trends in Juvenile Violence: A Report to the United States Attorney General on Current and Future Rates of Juvenile Offending. Northeastern University, Boston, MA. Greenfeld, Lawrence A., and Steven K. Smith (1999). American Indians and Crime. BJS Report NCJ 173386, U.S. Department of Justice, Washington, DC. Available at http://www.ojp.usdoj.gov/bjs/pub/pdf/aic.pdf. Illinois Criminal Justice Information Authority (1987). Trends and Issues: Criminal and Juvenile Justice in Illinois. Illinois Criminal Justice Information Authority, Chicago, IL. International Association of Chiefs of Police (1929). Uniform Crime Reporting. IACP, New York, NY. Kennedy, David M., Anne M. Piehl, and Anthony A. Braga (1996). "Youth Violence in Boston: Gun Markets, Serious Youth Offenders, and a Use-Reduction Strategy." Law and Contemporary Problems, 59, 1, 147-196. Lattimore, Pamela K., James Trudeau, K. Jack Riley, Jordan Leiter, and Steven Edwards (1997). Homicide in Eight American Cities: Trends, Context, and Policy Implications. Report NCJ 167262. National Institute of Justice, Washington, DC. Available at http://ncjrs.org/pdffiles/167262-1.pdf. Lott, John R., Jr. (1998). More Guns, Less Crime: Understanding Crime and Gun Control Laws. University of Chicago Press, Chicago, IL. Lott, John R., Jr., and David B. Mustard (1997). "Crime, Deterrence, and Right-to-Carry Concealed Handguns." Journal of Legal Studies, 26, 1, 1-68. Maltz, Michael D. (1972 [1999]). Evaluation of Crime Control Programs (monograph), National Institute of Law Enforcement and Criminal Justice, GPO: Washington. Internet version, published in 1999, found at http://www.uic.edu/~mikem/homepage.html Maltz, Michael D. (1976). "Secondary Analysis of the UCR: An Index of the Risk of Death Due to Robbery," Journal of Criminal Justice 4, 2, 153-56. Maltz, Michael D. (1977). "Crime Statistics: A Historical Perspective," Crime and Delinquency 23, 1, January, 32-40. Reprinted in Eric Monkkonen, Ed., Crime And Justice in American History, Meckler, 1990. Maltz, Michael D. (1998). "Visualizing Homicide: A Research Note," Journal of Quantitative Criminology, 14, 4, 397-410. Maxfield, Michael G. (1989). "Circumstances in Supplementary Homicide Reports: Variety and Validity." Criminology, 27, 4, 671-694. Poggio, Eugene C., Stephen D. Kennedy, Jan M. Chaiken, and Kenneth E. Carlson (1985). Blueprint for the Future of the Uniform Crime Reporting Program. Bureau of Justice Statistics and Federal Bureau of Investigation, U.S. Department of Justice, Washington, DC. Riedel, Marc (1990) "Nationwide Homicide Data Sets: An Evaluation of the Uniform Crime Reports and the National Center for Health Statistics Data." Chp. 9 in Doris Layton MacKenzie, Phyllis Jo Baunach, and Roy R. Roberg, Eds., Measuring Crime: Large-Scale, Long-Range Effects, State University of New York Press, Albany, NY. Sherman, Lawrence W., Denise Gottfredson, Doris MacKenzie, John Eck, Peter Reuter, Shawn Bushway, in collaboration with members of the Graduate Program (1997). Preventing Crime: What Works, What Doesn't, What's Promising. Department of Criminology and Criminal Justice, University of Maryland, College Park, MD. Snyder, Howard (1996). National Juvenile Violent Crime Trends, 1980-1994. National Center for Juvenile Justice, Pittsburgh, PA. Snyder, Howard (1999). "The Overrepresentation of Juvenile Crime Proportions in Robbery Clearance Statistics." Journal of Quantitative Criminology, 15, 2, 151-161. Stamp, Sir Josiah (1929). Some Economic Factors in Modern Life. P. S. King and Son, Ltd., London, UK. Williams, Kirk and Robert L. Flewelling (1987). "Family, acquaintance, and stranger homicide: Alternate procedures for rate calculations." Criminology, 25: 543-560. Zawitz, Marianne W., Ed. (1983). Report to the Nation on Crime and Justice: The Data. BJS Report NCJ 87068, U.S. Department of Justice, Washington, DC. Zawitz, Marianne W., Ed. (1988a). Report to the Nation on Crime and Justice: Second Edition. BJS Report, NCJ 105506, U.S. Department of Justice, Washington, DC. Zawitz, Marianne W., Ed. (1988b). Technical Appendix, Report to the Nation on Crime and Justice: Second Edition. BJS Report, NCJ 112011, U.S. Department of Justice, Washington, DC. Glossary BGP: Boston Gun Project, a multi-organization effort that served to reduce the number of youth homicides in Boston. BJA: Bureau of Justice Assistance, one of the agencies of the Office of Justice Programs that provides programmatic and financial assistance to State and local criminal justice agencies. BJS: Bureau of Justice Statistics, the statistical arm of the U.S. Department of Justice. CCSPD: Cook County (Illinois) Sheriff's Police Department. CHD: Chicago Homicide Dataset, a computer file of homicides maintained by the Illinois Criminal Justice Information Authority and available for downloading from NACJD. CIUS: Crime in the United States, the annual report on crime published by the FBI. CJIS: Criminal Justice Information Services Division of the FBI, responsible for the collection, analysis, and publication of crime, arrest, and other police-related data. DOJ: U.S. Department of Justice. FBI: Federal Bureau of Investigation, the investigative arm of the U.S. Department of Justice. IACP: International Association of Chiefs of Police, the organization that initiated the Uniform Crime Reporting Program. ICPSR: Inter-university Consortium of Political and Social Research, an organization established by colleges and universities to store political and social data files so they may be accessed by researchers for analysis. LEAA: the Law Enforcement Assistance Administration, an agency established by the Omnibus Crime Control and Safe Streets Act of 1968, in which was housed many of the agencies and functions now part of the Office of Justice Programs. NACJD: National Archive of Criminal Justice Data, a unit within the Inter-university Consortium of Political and Social Research that serves as a repository of criminal justice data funded by BJS and NIJ and housing data files produced by these agencies and their grantees/contractors, from which the files can be obtained. NCHS: National Center for Health Statistics, a statistical arm of the U.S. Department of Health and Human Services. NIBRS: National Incident-Based Reporting System, the crime data collection system that will replace the UCR system. NIJ: National Institute of Justice, the research arm of the U.S. Department of Justice. OJP: Office of Justice Programs, an office within DOJ that houses the Bureau of Justice Statistics, Bureau of Justice Assistance, National Institute of Justice, and other agencies that deal with State and local criminal justice agencies, issues and programs. ORI: Originating Agency Identifier, the identification number used by the FBI to identify police agencies PSS: Program Support Section, the section of the FBI's Criminal Justice Information Services Division that deals with the collection, analysis, and publication of crime data. SAC: Statistical Analysis Center, an agency established in many States during the 1970's, often with funding from BJS. SAS: Statistical Analysis System, computer software for data analysis. SHR: Supplementary Homicide Reports, a supplement to the Uniform Crime Reporting Program that collects information about each homicide incident, including victim and offender characteristics, the relation between victim(s) and offender(s), and incident circumstances. SPSS: Statistical Program for the Social Sciences, computer software for data analysis. UCR: Uniform Crime Reports or the Uniform Crime Reporting Program, the FBI program that collects, analyzes, and publishes crime, arrest and police personnel data from police agencies throughout the United States.