CRIME RATES

The interpretation of crime rates seems unproblematic, but those sociologists who study crime know that is not the case. Even simple counts of crime raise difficult issues. To imagine the difficulties consider two men at a bar. One of them makes a rude comment to the other and the other person shoves him. The person shoved strikes the other person on the jaw before the fight is broken up. The person hit on the jaw calls the police to the scene and wants to press charges, but the person shoved claims self-defense and that the person who shoved him is guilty of an assault. The two men may decide that the matter is not worth pressing and the police may agree. On the other hand, the police may file a report that results in an arrest. If a grand jury thinks the evidence warrants it, the case may go to trial. Even if the case reaches trial, the jury may acquit the accused.Other possible steps in the process exist, but determining whether a crime occurred is often difficult. The process contains many decision points. Will the individuals involved report the incident to the police? Will the police determine that a crime occurred? Now imagine that an interviewer asks one of these men about their history of criminal offending or victimization. Would the individual report this incident to the interviewer? Would one of the individuals admit to an interviewer that he committed an assault? Crime counts constitute an important element in the calculation of crime rates, but they are only one factor.

Rather than report simple crime counts, social scientists often calculate crime rates. Depending on the context, they may do this because crime rates: (1) allow for the comparison of crime patterns across groupings of different sizes, (2) make possible the comparison of the relative frequency of crime over time, or (3) help in the assessment of the risk of victimization. In terms of comparability of patterns across groupings of different sizes, simply reporting the number of crimes in cities with 100,000 or more residents produces unsatisfactory results. Forty homicides may not represent a large number for New York City, but twenty homicides represent a large number in Eugene, Oregon. The calculation of homicide rates per 100,000 residents for New York City and Eugene makes comparisons between the two cities easier. Similar calculations of rates for different years facilitate comparisons over time. The number of homicides may have doubled in Los Angeles from 1920 to 1990 while the rate per 100,000 residents has decreased. In terms of the risk of motor vehicle theft, a rate based on the number of motor vehicle thefts per 10,000 registered motor vehicles may be more helpful than the number of motor vehicles stolen.

Equation (1) represents the general formula for calculating a crime rate:

Each of the components in this formula: Base, Relevant Population Size, and Number of Incidents, represents an important decision point when calculating a crime rate. Determining the most appropriate measure for each component is not as straightforward as it might first appear.

Selecting a Base: If the base is 100,000, then the interpretation of the rate is the number of incidents per 100,000. With a base of 100 the interpretation is in terms of the number of incidents per 100 (a percent). Reports of crime rates often appear as rates per 100,000 or per 10,000. One reason for this is the relative infrequency of some crimes. For example, in 1997 the number of murders in the United States was 18,210 and the population of the United States was 268 million. If we choose 100 for the base, we find that the homicide rate per 100 is .0068 or .0068 percent. This figure challenges the intuition of most people. We should choose the base to make the rate easier to interpret. If we choose 100,000 for the base, we find that the homicide rate is 6.8 per 100,000; a much easier figure for most people to understand.

Relevant Population Size: The choice of the relevant population size depends upon the selection of an appropriate "relevant population." The Uniform Crime Reports (UCR) typically chooses the number of residents as the relevant population. This number is used to calculate crime rates for cities or for the United States as a whole in the UCR. The National Crime and Victimization Survey (NCVS) often uses the number of residents who are twelve years of age or older as the relevant population, since the survey only asks about crime incidents involving those in that age range. At other times households represent the relevant population as when victimization surveys calculate rates for households touched by crime.

Thus, the total number of people may not be the most appropriate "relevant population." In the case of rape incidents, it might be more appropriate to compute rape rates using women as the relevant population. For motor vehicle theft, the relevant population might be the number of registered motor vehicles. For commercial burglaries, the relevant population might be the number of commercial establishments. Social scientists should choose these populations carefully.

Number of Incidents: Determining the value for the numerator of the crime rate formula raises particularly difficult issues. The UCR, for example, employs a complicated set of rules for counting the number of incidents. For example, the UCR "hierarchy rule" ensures that for Part I crimes (the most serious street crimes) only the most serious crime is recorded. Thus, if a person breaks into a store, is confronted by the owner and threatens the owner with a gun, and then kills the owner, the offense is recorded as a homicide, but not as a burglary or robbery or assault. The "hotel rule" is invoked when a burglar breaks into several units in an apartment or hotel and only one burglary is counted. Researchers draw a distinction between an incident rate and a prevalence rate. This involves a distinction between the number of criminal incidents per some relevant population and the number of offenders per some relevant population. Using the number of offenders per some relevant population produces a prevalence rate.

Thus, many decisions help determine the most appropriate crime rate for a particular purpose. These decisions need to be considered when interpreting a particular rate or when attempting to compare one crime rate with another.

SOURCES OF CRIME DATA IN THE UNITED STATES

There are three major sources of crime data in the United States, and knowing the strengths and weaknesses of these data sources helps explain why crime rates derived from them often vary.

These three methods are similar to those used to collect crime data internationally: (1) official data collected by law enforcement agencies, (2) surveys of the victims of crimes, and (3) self-report studies based on offenders. Other sources also are available, but less commonly used (e.g., vital statistics, hospital records, and insurance records).

The Uniform Crime Reports. During the later half of the 1920s the International Association of Chiefs of Police (IACP) developed uniform crime definitions and counting rules that would allow it to collect more comparable data from different law enforcement agencies. The IACP initiated the UCR in January of 1930. During the latter part of 1930 the Bureau of Investigation took over the UCR program, and it has remained a function of that bureau (renamed the Federal Bureau of Investigation, or FBI).

The general definitions of the UCR Part I offenses appear in Table 1, but more detailed definitions appear in the Uniform Crime Reporting Handbook (FBI 1985). As an example, the Uniform Crime Reporting System defines burglary as the unlawful entry of any fixed structure, vehicle, or vessel used for a regular residence, industry, or business, with or without force, with the intent to commit a felony, or larceny. Without uniform definitions, jurisdictions in which state law defines theft from a storage-shed as a burglary would report such an incident as a burglary to the FBI. In another jurisdiction, breaking and entering might be required for an incident to be classified as a burglary. In such a jurisdiction an incident in which an offender entered a house through an open window in order to steal a stereo would be coded as unlawful entry with intent to commit a crime for the purposes of the state's law, but as a burglary for UCR purposes.

Participation in the UCR is voluntary. Local law-enforcement agencies send their data to the FBI or to state-level UCR programs that forward the data to the FBI. In the early years only the largest law enforcement agencies in the United States participated. But by the 1980s and 1990s, agencies representing about 97 percent of the population of the entire United States participated. The FBI reports two types of crime: Part I crimes and Part II crimes. For Part I crimes, data on crimes known to the police and on arrests are reported; for Part II crimes, only data on arrests are reported. Part I crimes include violent crimes (murder and non-negligent homicide, forcible rape, robbery, and aggravated assault), property crimes (burglary, larceny-theft, and motor vehicle theft), and arson. Each year the news media report crime rates derived from the FBI data and a summary of these data appear annually in Crime in the United States. This FBI publication provides crime rates for states, metropolitan statistical areas, counties, cities, other geographical subdivisions, and the nation as a whole.

Table 2 presents Part I crime data from the publication Crime in the United States—1997 (FBI 1998). The second column presents the number of incidents of a specific type, the third column reports the rate per 100,000 residents, and the final column indicates the change in the rate over a ten year period. In 1997, for example, 1,634,770 violent crimes were known to the police. The rate of violent crimes was 610.8 per 100,000 persons, which represents a decrease in the rate of violent crimes from 1988 to 1997 of 4.1 percent. Officially reported crime rates fell in all Part I categories from 1988 to 1997.

Crimes reported or discovered by the police and found through investigation to have occurred are labeled "crimes known to the police." A crime found not to have occurred is "unfounded," and does not appear as a crime known to the police. Although occasionally police discover a crime being committed (e.g., they happen upon a burglary in progress, or they arrest someone for resisting a police officer) they are highly dependent on citizen reports of criminal incidents.

The UCR also provides data on arrests for Part I and Part II crimes, and reports these crimes on the basis of the geographical divisions described above. Since these data are for arrests, the rates calculated from them should be labeled "arrest rates." Part II crimes include: other assaults (not including aggravated assaults), forgery and counterfeiting, fraud, embezzlement, stolen property (buying, receiving, and possessing), vandalism, weapons (carrying, possessing, etc.), prostitution and commercialized vice, sex offenses (except forcible rape and prostitution), drug abuse violations, gambling, offenses against family and children, driving under the influence, liquor law violations, drunkenness, disorderly conduct, vagrancy, curfew and loitering law violations, and runaways.

National Crime and Victimization Survey (NCVS). The NCVS data come from large surveys using probability samples of respondents. Since 1973 a national-level program has surveyed a probability sample of U.S. households. The U.S. Census Bureau conducts the survey, and during the 1990s the sample contained some 60,000 households and approximately 100,000 respondents. The results of this survey have been published annually since 1973 by The U.S. Department of Justice in Criminal Victimization in the United States.

The NCVS collects data on a much smaller number of crimes than the UCR. Since it is a survey of the victims of crime, it does not record homicides, but it does record forcible rapes and sexual assaults, robberies, aggravated assaults, simple assaults, burglaries, larceny-thefts, and motor-vehicle thefts. It also provides details about the victims of crime not available in the UCR. These include the household income of the victim, any injuries sustained by the victim, insurance coverage, length of any hospitalization due to injuries, what protective measures the victim employed, and so on.

Careful field-testing preceded the initiation of data collection for NCVS in 1973 and before any changes made to its design since that time. An early decision involved the length of the reference period (the period the interviewers referred to when they asked whether victimizations occurred during the past period). Researchers chose a six-month reference period on the basis of the costs of doing more frequent surveys and studies that showed the effect of memory decay on remembering whether an event occurred. Respondents tend to telescope events into the reference period, and this can cause substantial increases in the number of victimizations reported during the past six months. To overcome this problem, each respondent remains in the sample for three years with interviews repeated every six months. Victimizations reported to the interviewer the first time a respondent is interviewed do not contribute to estimates of the crime rates. That first interview and subsequent interviews provide bounds for the six-month reference period. The interviewer asks about victimizations occurring since the last interview and has a list of incidents reported in the last interview to make sure that those incidents are not telescoped into the reference period.

To estimate rates for the entire nation from the NCVS data, researchers use the samples to estimate the number of victimizations occurring in the entire nation. They then use these estimates to calculate victimization rates. The derived estimates often contain large sampling errors, since they are based on only a "small" sample of the total population of the United States. These large sampling errors are greatest for the crimes least frequently reported by the respondents, for example, victimization rates for rapes involve more sampling error than rates for aggravated assault (O'Brien 1986).

Self-Reports. Unlike the UCR and NCVS, self-report studies refer to a collection of studies conducted by different researchers using a variety of methodologies. The basic methodology, however, involves sampling respondents and questioning them about criminal or delinquent acts. Although some isolated self-report studies occurred in the 1940s, the technique gained popularity with the work of Short and Nye (1957, 1958).

Perhaps the most ambitious self-report study is the National Youth Survey, or NYS (Elliott et al. 1983). This survey involves a national panel sample. When first instituted in 1976, the panel members were aged eleven to seventeen. Estimates of national-level crime rates for the age groups covered in this survey can be computed because the sample of respondents is a national one. These rates, unlike UCR- and NCVS-based rates, result from respondents answering questions about their offenses.

The NYS, and other self-report surveys, calculate two types of crime rates: "prevalence rates" and "incidence rates." To calculate these crime rates, the NYS uses the general formula represented in Equation (1). The prevalence rate uses the number of respondents who claimed they committed an offense for the numerator, the number of respondents for the relevant population size, and (we choose) 100 for the base. This yields the percentage of respondents who have committed a particular act. In the NYS in 1980, when the respondents were fifteen to twenty-one years old, the prevalence rate for robbery was 2. The incidence rate uses the number of times individuals claimed they committed a particular act as the numerator, the number of respondents as the relevant population size, and typically uses 100, 1000, or 10,000 for the base. We again use 100 for the base and find that in the 1980 NYS the incident rate for robbery was 10. Thus, there was an average of ten robberies per 100 respondents in this age group in 1980, with only two out of 100 respondents claiming involvement in a robbery.

As with the NCVS, self-report studies depend upon the accuracy of respondent reports. Therefore, issues such as telescoping, recall, and embarrassment are consequential for self-reports of criminal activities. In comparison to the UCR and NCVS, many of the crimes reported in self-report surveys are trivial crimes or not crimes at all (e.g., lying to parents or cheating on a test). The sample sizes, even in the largest of such surveys, are not nearly as large as sample sizes in the NCVS. Therefore, given the relative infrequency of serious crimes, the rates for serious crimes are not very precisely measured by self-report surveys; that is, the estimates have large standard errors. Nonetheless, crime rates derived from self-report studies provide much of the data used to investigate the determinants of individual criminal offending.

SOME PROBLEMS WITH UCR, NCVS, AND SELF-REPORT DATA

Using information from official data (UCR), victimization surveys (NCVS), and self-report surveys helps us to gauge the accuracy of crime rates. Crime rates vary depending upon whether they derive from UCR, NCVS, or self-report data. To understand why crime-rate estimates vary depending upon their source, we briefly outline some problematic aspects of these three data sources.

UCR Data. The most obvious problem with UCR data involves underreporting of crimes to the police. The NCVS interviewers ask respondents whether they reported a particular incident to the police. In 1994, 78 percent of the respondents who reported a motor vehicle theft to the interviewer said they reported that crime to the police, 50 percent of the burglaries were reported, 50 percent of the robberies, and only 32 percent of rapes and sexual assaults (U.S. Department of Justice 1997).

Once an incident comes to the attention of the police a number of factors influence whether the incident is recorded as a crime or whether an arrest is made. There may be organizational pressures to raise or lower the crime rate (e.g., Bell 1960; Chambliss 1984; DeFleur 1975; McCleary et al. 1982; Selke and Pepinsky 1982; Sheley and Hanlon 1978; Wheeler 1967). The professional styles of particular police departments may affect the recorded crime rates (Beattie 1960; Skogan 1976; Wilson 1967, 1978). The interactions between officers and offenders also help determine the classification of an incident as a crime and whether an arrest is made (Black 1970; Smith and Visher 1981; Visher 1983). Given these and other problems, it is not surprising that crime rates recorded by the UCR are known to contain substantial errors.

NCVS Data. The NCVS may well provide a more accurate picture of crime rate trends and comparative crime rates than the UCR. As Sellin's (1931, p. 346) dictum suggests, "the value of a crime for index purposes decreases as the distance from the crime itself in terms of procedure increases." Further, the NCVS standardizes the procedures it uses in compiling crime incidents. When the NCVS changes its procedures, careful methodological studies evaluate the changes.

Although the incidents reported to NCVS interviewers need not depend upon a respondent contacting the police, the police responding to the contact, and eventually police recording of the incident as founded, respondents do need to respond appropriately in an interview situation in order for an incident to be recorded as a victimization in the NCVS. In this context, note that the interview situation is a social interaction. The interviewers have little to offer respondentd for their time. Respondents may be reluctant to share embarrassing information with the interviewer, such as a fight at a bar, an assault by a relative, or a sexual assault. As noted earlier, sometimes respondents may forget about incidents or telescope the incidents into or out of the reference period.

Turner (1972) used "reverse record" checks to investigate 206 cases of robbery, assault, and rape found in police records. Interviewers interviewed these "known victims" using an NCVS-like technique. Only 63.1 percent of these "known incidents" were reported to interviewers. The percentage reporting these incidents to the interviewer was strongly related to the relationship of the offender to the victim. Respondents reported 76.3 percent of the incidents involving a stranger; 56.9 percent of the incidents involving known offenders; and only 22.2 percent of the incidents involving a relative.

Not all of this underreporting of incidents to the interviewers is attributable to embarrassment. Turner (1972) found a strong relationship between the number of months between the interview and the incident and the respondents' reporting of the incident to the interviewer. Respondents recalled 69 percent of the incidents occurring one to three months before the interview, 50 percent of those occurring four to six months before, 46 percent of those occurring six to nine months before, and only 30 percent of those occurring ten to twelve months before.

Self-Report Data. Like UCR and NCVS data, self-report data contain their own weaknesses. As with the NCVS data, the survey research situation raises questions of honesty, forgetting, bounding the reference period, and so on. Some problems are more severe than those encountered by the NCVS. One of these problems is sample size; even the best-financed surveys—like the National Youth Survey (Elliot et al. 1983)—have samples of less than 2,000. Such small samples almost guarantee large sampling errors associated with the resulting crime rates (especially for serious crimes). Small sample size further limits the usefulness of these data for conducting regional analyses (e.g., estimating the crime rates for states or cities). Similarly, it limits the accuracy of these estimates for comparing the crime rates for groups such as Hispanics or Asian females. Unlike the NCVS, many self-report surveys do not involve panels in which respondents are reinterviewed over an extended period of time (the NYS is an exception) so that the interviews cannot be bounded. Often the response rates are quite low. In the NYS (Elliot et al. 1983), a sample of 2,360 eligible youth originally were selected and of these 73 percent agreed to participate in the first wave of data collection in 1976. By 1980 the sample size dropped to 1,494 or 63 percent of the original sample. This contrasts with initial response rates well in excess of 90 percent for the NCVS.

Two other issues need mention. First, self-report studies concentrate on relatively trivial behaviors, such as lying to parents, defying authority, or cheating on tests. One reason for concentrating on such behaviors is that they are more commonly reported. Given the small sample sizes typical of self-report studies, these behaviors may be more reliably measured. Second, some studies do not accurately gauge the frequency of "delinquent behaviors." They use response sets such as "no," "once or twice," or "several times." Experienced researchers now ask about a wide range of behaviors and more specifically ask about the frequency of these behaviors. In the NYS, Elliot et al. (1983) ask about "arson," "prostitution," and "physical threat for sex" as well as "skipped class" and "didn't return change." They use response categories such as "2–3 times a day," "once a day," "2–3 time a week," "once a week," "once every 2–3 weeks," and "once a month," as well as asking the exact number of times respondents offended during the past year.

COMPARISONS OF CRIME RATES GENERATED FROM UCR, NCVS, AND SELF-REPORTS

Fortunately, the problems outlined above do not mean that crime rates based on UCR, NCVS, and self-reports are not useful. Homicide rates, for example, are probably reasonably accurate in terms of absolute rates and in terms of homicide trends over the past sixty-five years. They also are appropriate for comparisons of homicide rates across large cities. When we judge the usefulness and accuracy of crime rates, the type of crime considered is important (e.g., homicide or rape). Further, even if a crime rate is inaccurate in terms of the absolute amount of crime it represents, it may be an accurate measure of the relative amount of that crime in different areas or across time. This would be the case if exactly half or some other proportion of the aggravated assaults were reported in different jurisdictions (or across time). Then we could perfectly compare the relative amount of crime across jurisdictions (or crime trends over time). If only approximately the same proportion of cases were reported across jurisdictions, then comparisons of the relative amount of crime would only be approximate.

Measuring Absolute Crime Rates. The most accurately measured UCR Part I crime almost certainly is homicide. The reasons for this include the seriousness of the crime and the physical evidence it produces (e.g., a body, weapon, or missing person), which make it unlikely that once detected this crime will be ignored. Cantor and Cohen (1980) compared homicide rates based on the U.S. Department of Health, Education, and Welfare's Annual Vital Health Statistics Report with those from the UCR for the years 1935–1975 and found a very close correspondence. Using a different approach, O'Brien (1996) found evidence that fluctuations in the homicide rate closely paralleled fluctuations in the rate of other violent crimes from 1973 to 1992.

Motor vehicle theft is the other Part I crime that appears to be well estimated in terms of absolute rates. In 1994, for example, the NCVS estimated 1,764,000 motor vehicle thefts based on respondents' answers, while the UCR reported 1,539,100 motor vehicle thefts known to the police. Thus the NCVS estimate exceeded the UCR estimate by only 28 percent. When NCVS respondents report a motor vehicle theft they are asked if they reported the theft to the police. Based on their responses one would calculate that the police receive reports of only 1,379,448 motor vehicle thefts, which is fewer motor vehicle thefts than appear in the UCR records. This probably occurs because motor vehicle thefts from private citizens constitute only 80 to 85 percent of the total number of such thefts known to the police (Biderman and Lynch 1991). For the remaining Part I crimes covered by both the UCR and NCVS (personal robbery, aggravated assault, residential burglary, and rape), the NCVS and UCR data show greater differences in absolute rates. The smallest ratio of NCVS incidents to UCR incidents is 1.62 for rape and the largest 3.01 for residential burglary.

Difficulties arise when comparing the self-report rates to UCR and NCVS rates, because self-report studies typically only involve a sample of young people. The estimated crime rates for these young people, however, are so high that we may conclude that the rates generated by self-reports are far greater than those generated by either the NCVS or the UCR. The 1976 NYS produced an estimated incident rate for eleven to seventeen year-olds for aggravated assault of 170 per 1000 and for robbery a rate of 290 per 1000. This compares with rates for those twelve year-olds and over of 7.90 and 6.48 based on the NCVS and rates for all ages of 2.29 and 1.96 based on the UCR. Some part of this discrepancy results from differences in the age groups compared, but not all of it. In 1976 eleven to seventeen year-olds comprised 13.38 percent of the population. Thus, even if it were assumed that no assaults or robberies were committed by other age groups, the aggravated assault rate based on the NYS would be [.1338 × 170 =] 22.75 and the rate for robbery would be [.1338 × 290 =] 38.80.

Measuring Relative Crime Rates. If absolute rates are well measured, then relative rates of crimes across cities or other units probably are well measured. This means that both homicide rates and motor vehicle theft rates should be good measures of the relative rate of crime. In the mid-1970s the NCVS conducted victimization studies with reasonably large size samples; 10,000 or more households containing some 22,000 respondents, in each of twenty-six large U.S. cities. The same cities, of course, reported UCR crime rates for Part I crimes. This provided the opportunity to compare the consistency of these estimates in terms of the relative crime rates measured by the NCVS and the UCR by correlating the two crime-rate measures across the twenty-six cities. The closer these correlations are to 1.00, the greater the degree to which cities with relatively high rates (low rates) on one of the measures have relatively high rates (low rates) on the other. Across these cities, the correlation of UCR and NCVS motor vehicle theft rates was .90 or higher (Nelson 1978, 1979; O'Brien, Shichor, and Decker 1980). The correlations were close to zero for rape, negative for aggravated assault, and positive and moderately strong for burglary and robbery. This indicates that for burglary and robbery the relative rates of crime in these cities are similar whether the UCR or NCVS measures are used, but they are not similar for rape and aggravated assault.

Overall, the use of homicide rates and motor vehicle theft rates to measure the relative amount of these crimes across cities is supported by the above findings. The use of burglary and robbery rates receives weaker support, and the use of rape and aggravated assault rates to compare the relative amount of these crimes across cities does not receive support.

Gender and Race Composition of Offenders. Hindelang (1978, 1979) addressed the issue of whether the UCR and victim reports (NCVS) produce similar percentages of offenders identified as African American or white or as male or female. O'Brien (1995, 2000) replicated these finding using more recent data. These comparisons indicate that for those crimes in which the victim and offender come into contact (which enables the victim to identify the race and sex of the offender), the percentages based on the UCR and NCVS are fairly similar. For example, O'Brien (2000) aggregates data for the years 1992 to 1994 and finds that the percentages of males involved in these crimes, according to the UCR and NCVS, never differ by more than 3 percent. For race of the offender, rape produces the largest difference with victims stating that 33.2 percent of the offenders are African American and the UCR arrest data indicating that 42.7 percent of those arrested are African American. For robbery, aggravated assault, and simple assault the differences do not exceed 5 percent.

Self-report data at one time were considered to be inconsistent with UCR data, which indicated that males committed crimes at a much higher rate than females and that African Americans also committed crimes at a much higher rate than whites. These large differences between males and females and African Americans and whites, however, are not found when self-report studies ask about more serious crimes and ask more carefully about the frequency of offending (Elliot and Ageton 1980; Hindelang, Hirschi, and Weis 1979).

These comparisons only touch on the issue of the appropriate uses of UCR, NCVS, and self-report data. The appropriate uses depend upon the type of crime and the type of comparison. (For helpful information on which to judge the appropriate uses of these data see Biderman and Lynch 1991; Gove Hughes and Geerken 1985; O'Brien 1985, 2000).

INTERNATIONAL DATA

Many nations collect data from law enforcement agencies, survey victims, and conduct self-report surveys. Difficulties arise, however, when comparing these data, since the definitions of crimes and the counting rules differ from nation to nation. In this section we note some of these problems and cautiously present crime-rate data for two nations. In addition, for a larger number of nations, we present homicide rates.

Perhaps the most comparable large national victimization surveys are the NCVS (conducted in the United States) and the British Crime Survey (conducted in England and Wales). But even in this case, there are important differences—the British Crime Survey (BCS) uses a twelve-month reference period and unbounded interviews, while the NCVS uses a six-month reference period and bounded interviews. The former difference should decrease the estimated victimization rates in the BCS due to "memory decay," but the latter should increase these rates due to "telescoping." The BCS does not ask about crimes involving victims under the age of sixteen, while the NCVS does not ask about crimes involving victims under age twelve.

Further problems arise with the definitions of offenses. In the United States, for example, aggravated assaults include attempted murders and almost any assault in which a weapon is used or assaults for the purpose of inflicting severe or aggravated injury. In England attempted murders receive separate treatment while the assaults most similar to Part I aggravated assaults are "woundings." Woundings require some kind of cut or wound where the skin or a bone is broken, or medical attention is needed, whereas common assault occurs if the victim is punched, kicked or jostled, with negligible or no injury.

Since 1981 definitions of rape have changed substantially in England. Until 1981, to be classified as a rape an incident required a male offender and female victim and the penetration of the vagina by the penis. Husbands could not be convicted of raping their wives, and males under age fourteen could be convicted of a rape. By 1996 rapes could involve offenders as young as ten, spousal victims, male victims, and anal intercourse. In the United States, the UCR definition of rape includes male victims, male or female offenders, spousal victims, anal intercourse, and other sexual acts. With these nontrivial problems in mind, we compare crime rates (based both on surveys and police records) in England and the United States.

Table 3 presents victimization rates for robbery, assault, burglary, and motor vehicle theft in the United States and England (based on the NCVS and the BCS) for the years 1981 and 1995 (Langan and Farrington 1998). Perhaps the most surprising result is that the estimated victimization rates in England are higher than in the United States for these crimes in 1995. The surveys also indicate substantial increases in robbery, assault, burglary, and motor vehicle theft from 1981 to 1995 in England and decreases for all but motor vehicle theft in the United States. The cross-country comparisons of these absolute rates may be more problematic than over-time comparisons within the countries—given that the victimization survey methods and the definitions of these four crimes changed little in these countries over this period.

Table 4 presents crime rates based on police records (Langan and Farrington 1998). In this

table, we report rates per 100,000 residents because of the relatively low rates for homicide and rape. The rates in Table 4 reinforce the patterns reported in Table 3, which showed a substantial increase in crime rates from 1981 to the mid-1990s in England for robbery, assault, burglary, and motor vehicle theft. Table 4 also shows higher rates for these crimes (except for robbery) in England than in the United States in 1996. Two new crimes appear in this table: Homicide, for which the rate in the United States is several times higher than in England, for both 1981 and 1996 (most criminologists agree that homicide rates in the United States are very high by international standards) and rape, which shows a dramatic increase from 4.19 to 21.77 per 100,00 in England (this results, at least in part, from changes in the legal definition of rape that occurred in England between 1981 and 1996).

Although we could attempt other comparisons, most such comparisons involve even greater difficulties than those between the United States and England. Some countries do not differentiate between simple and aggravated assaults; others do not record any but the most severe assaults when the assault occurs within the family. As noted, legal definitions of rape may differ widely (even over time within the same country), reports of burglaries may require breaking and entering into a business or residence, only entering, or even breaking into an outbuilding.

Homicide represents the crime with the most consistent definition across countries: the intentional taking of another person's life without legal justification. Even here some countries may not separate justifiable from unjustifiable homicides.

The legal justifications for killing a spouse or even a person committing a crime may vary from country to country, as may the reporting and recording of homicides. Still, the data for this crime rate almost certainly are more accurate for international comparisons than those for rape, robbery, burglary, assault, theft, or arson.

Table 5 presents data for 1990 from the United Nations (United Nations Crime and Justice Network 1998) on intentional homicide rates per 100,000 for a selected set of countries. The rate for the United States was taken from the UCR (FBI 1991). Note the wide range of homicide rates, with Barbados and the United States with relatively high rates and Austria, Botswana, Denmark, Japan, Norway, and Spain with relatively low rates. There probably are real differences between these two sets of countries in terms of homicide rates. It is less clear that smaller differences across countries with rather different legal systems, levels of development, and cultural histories, represent differences in the unjustifiable intentional taking of another person's life. To reiterate the difficulty in making cross-national comparisons, we note that INTERPOL's international crime statistics come with the warning: "the information given is in no way intended for use as a basis for comparisons between different countries."

CONCLUSIONS

At the beginning of this article we noted that the computation of a crime rate might seem straightforward, but designating the most appropriate term for each component in equation (1) often is

difficult. Designating the base represents the easiest decision. This choice does not greatly change the interpretation of the rate, although it may make the interpretation more or less intuitive for the reader. Choice of the appropriate relevant population size is more difficult; for example, basing rape rates on the number of females rather than the number of people or the motor vehicle theft rate on the number of registered motor vehicles rather than the number of people. Determining the number of incidents constitutes the most problematic component of calculating crime rates. Here, the differing definitions of crimes, varying counting rules, response rates, rates of reporting incidents to the police or interviewers, response categories, bounding of interviews, memory decay, police discretion in recording crimes, and so on, can greatly affect the estimated crime rate. These factors differentially affect estimates of the number of incidents based on self-reports, UCR, and NCVS data. Even so, within the United States for certain crimes and comparisons, data from these different sources lead to similar conclusions. Comparisons of crime rates across nations, with the possible exception of homicide, is extremely risky.

references

Beattie, Ronald H. 1960 "Criminal Statistics in the United States—1960." Journal of Criminal Law, Criminology, and Police Science 51:49–65.

Bell, Daniel 1960 "The Myth of Crime Waves." In Daniel Bell, ed., The End of Ideology: On the Exhaustionof Political Ideas in the Fifties. New York: Free Press.

Biderman, Albert D., and James P. Lynch 1991 Understanding Crime Incidence Statistics: Why the UCR Diverges From the NCS. New York: Springer-Verlag.

Black, Donald J. 1970 "The Production of Crime Rates." American Sociological Review 35:733–748.

Cantor, David, and Lawrence E. Cohen 1980 "Comparing Measures of Homicide Trends: Methodological and Substantive Differences in Vital Statistics and Uniform Crime Report Time Series (1935–1975)." Social Science Research 9:121–145.

Chambliss, William J. 1984 "Crime Rates and Crime Myths." In William J. Chambliss, ed., Criminal Law inAction. New York: John Wiley.

Defleur, Lois B. 1975 "Biasing Influences on Drug Arrest Records: Implications for Deviance Research." American Sociological Review 40:88–103.

Elliott, Delbert S., and Suzanne S. Ageton 1980 "Reconciling Race and Class Differences in Self-Reported and Official Estimates of Delinquency." AmericanSociological Review 45:95–110.

——, David Huizinga, Brian A. Knowles, and Rachelle J. Canter 1983 The Prevalence and Incidence of Delinquent Behavior: 1976–1980: National Estimates of Delinquent Behavior by Sex, Race, Social Class and OtherSelected Variables. Boulder, Colo.: Behavioral Research Institute.

Federal Bureau of Investigation 1985 Uniform CrimeReporting Handbook. Washington, D.C.: U.S. Government Printing Office.

——1991 Crime in the United States—1990. Washington, D.C.: U.S. Government Printing Office.

——1998 Crime in the United States—1997. Washington, D.C.: U.S. Government Printing Office.

Gove, Walter R., Michael Hughes, and Michael Geerken 1985 "Are Uniform Crime Reports a Valid Indicator of the Index Crimes? An Affirmative Answer with Minor Qualifications." Criminology 23:451–491.

Hindelang, Michael J. 1978 "Race and Involvement in Common Law Personal Crimes." American Sociological Review 43:93–109.

——1979 "Sex Differences in Criminal Activity." Social Problems 27:143–156.

——, Travis Hirschi, and Joseph G. Weis 1979 "Correlates of Delinquency: The Illusion of Discrepancy Between Self-Report and Official Measures." American Sociological Review 44:995–1014.

Langan, Patrick A., and David Farrington 1998 Crimeand Justice in the United States and In England andWales, 1981–96. Washington, D.C.: U.S. Department of Justice.

McCleary, Richard, Barbara C. Nienstedt, and James M. Erven 1982 "Uniform Crime Reports as Organizational Outcomes: Three Time Series Experiments." Social Problems 29:361–372.

Nelson, James F. 1978 "Alternative Measures of Crime: AComparison of the Uniform Crime Report and the National Crime Survey in 26 American Cities." Presented at the American Society of Criminology, Dallas, Tex.

——1979 "Implications for the Ecological Study of Crime: A Research Note." In William H. Parsonage, ed., Perspectives on Victimology. Beverly Hills, Calif.: Sage.

O'Brien, Robert M. 1985 Crime and Victimization Data. Beverly Hills, Calif.: Sage.

——1986 "Rare Events, Sample Size, and Statistical Problems in the Analysis of NCS (National Crime Survey) City Surveys." Journal of Criminal Justice 14:441–448.

——1996 "Police Productivity and Crime Rates: 1973–1992." Criminology 34:183–207.

——2000 "Crime and Victimization Data." In Joseph F. Sheley, ed., Criminology. New York: Wadsworth.

——, David Shichor, and David L. Decker 1980 "An Empirical Comparison of the Validity of UCR and NCS Crime Rates." Sociological Quarterly 21:301–401.

Selke, William L., and Harold E. Pepinsky 1982 "The Politics of Police Reporting in Indianapolis, 1948–1978." Law and Human Behavior 6:327–347.

Sellin, Thorsten 1931 "The Bias of a Crime Index" Journal of the Institute of Criminal Law and Criminology 22:335–356.

Sheley, Joseph F., and John J. Hanlon 1978 "Unintended Effects of Police Decisions to Actively Enforce Laws: Implications for Analysis of Crime Trends." Contemporary Crises 2:265–275.

Short, James F., and F. Ivan Nye 1957 "Reported Behavior as a Criterion of Deviant Behavior." Social Problems 5:207–213.

——1958 "Extent of Unrecorded Juvenile Delinquency: Tentative Conclusions." Journal of CriminalLaw and Criminology 49:296–302.

Skogan, Wesley G. 1976 "Crime and Crime Rates." In Wesley G. Skogan, ed., Sample Surveys of the Victims ofCrime. Cambridge, Mass.: Ballinger.

Smith, Douglas A., and Christy A. Visher 1981 "Street Level Justice: Situational Determinants of Police Arrest Decisions." Social Problems 29:167–177.

Turner, Anthony G. 1972 The San Jose Methods Test ofKnown Crime Victims. Washington, D.C.: U.S. Government Printing Office. National Criminal Justice Information and Statistics Service, Law Enforcement Assistance Administration.

United Nations Crime and Justice Information Network 1998 http://www.ifs.unvie.ac.at/uncjin/mosaic/at.inthom/txt.

U.S. Department of Justice 1997 Criminal Victimizationin the United States, 1994. Washington, D.C.: U.S. Government Printing Office.

Visher, Christy A. 1983 "Gender, Police, Arrest Decisions, and Notions of Chivalry." Criminology 21:5–28.

Wheeler, Stanton 1967 "Criminal Statistics: A Reformulation of the Problem." Journal of CriminalLaw, Criminology, and Police Science 58:317–324.

Wilson, James Q. 1967 "The Police and Delinquents in Two Cities." In Stanton Wheeler and H. M. Hughes, eds., Controlling Delinquents. New York: John Wiley.

——1978 Varieties of Police Behavior: The Managementof Law and Order in Eight Communities. Cambridge, Mass.: Harvard University Press.

Robert M. O'Brien