U.S. Department of Justice Office of Justice Programs Bureau of Justice Statistics Cross-National Studies in Crime and Justice September 2004, NCJ 200988 --------------------------------------------------------------- This file is text only without graphics and many of the tables. A Zip archive of the tables in this report in spreadsheet format (.wk1) and the full report including tables and graphics in .pdf format are available from: http://www.ojp.usdoj.gov/bjs/abstract/cnscj.htm ---------------------------------------------------------------- Edited by David P. Farrington, Patrick A. Langan, and Michael Tonry Contributors Marcelo F. Aebi, C. J. H. Bijleveld, Carlos Carcach, Lars Dolm‚n, David P. Farrington, Mark J. Irving, Darrick Jolliffe, Martin Killias, Philippe Lamon, Patrick A. Langan, Catrien Paul R. Smit, David J. Smith, Michael Tonry, Brandon C. Welsh, P.-O. H. Wikstr"m Lawrence A. Greenfeld Director Report of work performed under BJS Cooperative Agreement No. 2000-BJ-CX-0001 awarded to the Castine Research Corporation, P.O. Box 110, Castine, Maine 04421. Contents of this document do not necessarily reflect the views or policies of the Bureau of Justice Statistics or the U.S. Department of Justice. The U.S. Department of Justice authorizes any person to reproduce, publish, translate, or otherwise use all or any parts of the material in this publication with the exception of those items indicating they are copyrighted or printed by any source other than the authors. Carolyn C. Williams produced and edited the report under the supervision of Tom Hester. Jayne Robinson administered final production. An electronic version of this report may be found on the Internet at: Http://www.ojp.usdoj.gov/bjs/ Contents 1. Introduction David P. Farrington, Patrick A. Langan, Michael Tonry, and Darrick Jolliffe 2. England and Wales David P. Farrington and Darrick Jolliffe 3. United States Patrick A. Langan 4. Australia Carlos Carcach 5. Canada Brandon C. Welsh and Mark J. Irving 6. Netherlands Catrien C. J. H. Bijleveld and Paul R. Smit 7. Scotland David J. Smith 8. Sweden P.-O. H. Wikstr"m and Lars Dolm‚n 9. Switzerland Martin Killias, Philippe Lamon, and Marcelo F. Aebi Introduction In 1981 there were 88 residential burglaries per 1,000 households in the United States, compared with 41 per 1,000 households in England (including Wales). Why was the U.S. residential burglary rate twice as high as the English burglary rate? Ten years later the U.S. rate had decreased to 54 per 1,000 households, but the English rate had increased to 68 per 1,000 households (Farrington, Langan, and Wikstr"m, 1994). Why was residential burglary decreasing in the U.S., and why was it increasing in England? In 1981 it was estimated that 15 in every 1,000 U.S. burglary offenders were convicted, compared with 28 in every 1,000 English burglary offenders. Why was the probability of conviction for burglary twice as high in England? Ten years later the U.S. conviction rate had increased to 19 per 1,000 offenders, while the English rate had decreased to 10 per 1,000 offenders (Farrington and others, 1994). Why was the probability of conviction increasing in the U.S., and why was it decreasing in England? It is extremely difficult to explain trends in national crime rates and in the probability and severity of legal punishment, and differences between countries. There are enormous problems of comparability over time and between countries, in laws, measurement methods, recording practices, and macrosocial cultural and political factors. However, the first step in moving towards explanations is to obtain comparable information about crime and punishment over time, and the main aim of this report is to present such comparable information for six serious crimes in eight countries between 1981 and 1999. While there have been many previous attempts to collect and present cross-national data on crime and punishment, we believe that our authors have made the most comprehensive and sophisticated efforts so far to present comparable data. Method The six serious crimes that are studied are: 1. Residential burglary 2. Vehicle theft 3. Robbery 4. Serious assault 5. Rape 6. Homicide. The eight countries are: 1. England (and Wales) 2. United States 3. Australia 4. Canada 5. Netherlands 6. Scotland 7. Sweden 8. Switzerland The information about crime and justice covers (for each type of crime): 1. Number of crimes committed (according to a national victimization survey) 2. Number of crimes reported to the police (according to a national victimization survey) 3. Number of crimes recorded by the police 4. Average number of offenders committing each crime 5. Number of persons convicted 6. Number of persons sentenced to custody 7. Average sentence length 8. Average time served These quantities are compared with the population in each country, and linking probabilities are estimated (such as, the probability of an offender being convicted, the probability of a convicted person being sentenced to custody). Each author has written a chapter following the same template. It is essential to know the average number of offenders committing each crime in order to relate crimes committed to persons convicted. For example, if three persons jointly commit one crime, this can lead to three persons being convicted. Therefore, in calculating the probability of an offender being convicted, it is important to divide the number of persons convicted (in this example, three) by the number of offenders (that is, the number of offender-crime pairs, which in this example is three), not by the number of crimes (in this example, one). Because of problems of comparability, we have not attempted to collect data on all possible stages of the criminal justice system or on all possible sentences. For example, while U.S. national data on arrests have been published annually for many years (such as FBI, 2002), national arrest data were not collected in England until 1999. We focused on convictions and custody because the problems of comparability were least daunting in these cases. The information in this report represents basic data that any theory of crime or criminal justice should be able to explain. For six serious crimes in eight countries between 1981 and 1999, the following key questions are addressed: 1. How is the crime rate changing over time? 2. Is the probability of a victim reporting a crime to the police increasing or decreasing over time? 3. Is the probability of the police recording a crime that is reported to them increasing or decreasing over time? 4. How is the conviction rate changing over time? 5. Is the probability of an offender being convicted increasing or decreasing over time? 6. Is the probability of a convicted offender being sentenced to custody increasing or decreasing over time? 7. How is the average sentence length changing over time? 8. How is the average time served changing over time? 9. Is the average time served per offender increasing or decreasing over time? The average time served per offender is a summary measure of punitiveness, obtained by combining the probability of an offender being convicted, the probability of an offender being sentenced to custody following a conviction, and the average time served per custodial sentence. This report aims to document changes in these quantities over time in each country. The challenge to criminologists is to explain why there are changes over time and why there are differences between countries both in the absolute values of these quantities and in trends over time. The eight countries were chosen because they had repeated large-scale national victim surveys, together with adequate criminal justice data, between 1981 and 1999. It was decided that the sample size in the International Crime Victims Survey (ICVS: van Kesteren, Mayhew, and Nieubeerta, 2000) of about 2,000 per country was insufficient for our purposes (especially for estimating linking probabilities for particular crimes). Three other countries were considered for inclusion in our analyses, namely Germany, Italy and Finland. However, analyses for Germany were enormously complicated by its reunification in the middle of the Study time period, the victim survey data for Italy proved to be inadequate, and we did not succeed in recruiting a collaborator in Finland who was able and willing to carry out the necessary analyses. Previous research This chapter is a continuation of previous work, which began with David Farrington's presidential address to the British Society of Criminology in January 1990, attempting to link national-level data in England (including Wales), to estimate crime-specific numbers flowing through the criminal justice system at each stage, from crimes committed to crimes reported to the police, crimes recorded by the police, offenders convicted, offenders sentenced to custody, average sentence length, and average time served, in 2 years (1981 and 1987). The first year that it was possible to make these estimates in England was 1981, which was the first year of the national victim survey -- the British Crime Survey or BCS. Ideally, a longitudinal study is needed, beginning with offenses and tracking offenders through the different stages of the criminal justice system, using a unique identification number for each offender at each stage. Unfortunately, national-level data tracking individual offenders across the different stages of the criminal justice system are not available in England or in most other countries. However, aggregate national data are available for each of the stages separately (such as crimes committed, persons convicted, persons sentenced to custody). These separate counts do not arise from tracking the same individuals across stages, but they permit reasonably accurate estimates of the flow of offenders from one stage to the next. David Farrington and Patrick Langan compared trends over time in England and the United States. The first publication (Farrington and Langan, 1992) estimated numbers flowing through the criminal justice system in both England and the United States and compared trends over time (between 1981 and 1987 in England, between 1981 and 1986 in the United States). David Farrington and Per-Olof Wikstr"m then compared trends over time in England and Sweden. The second publication (Farrington and Wikstr"m, 1993) estimated numbers flowing through the criminal justice system in England and Sweden and compared trends between 1981 and 1987. All three countries were then compared in a third publication (Farrington and others, 1994), which extended the previous analyses to between 1981 and 1991 for England and Sweden and between 1981 and 1990 for the United States. One problem with the first three analyses is that they reported changes between two widely separated time points rather than trends over time, making it difficult to know precisely when changes occurred and how they might be explained. The fourth publication (Langan and Farrington, 1998) calculated all numbers and probabilities for 7 years in the United States (1981, 1983, 1986, 1988, 1990, 1992 and 1994) and for 6 years in England (1981, 1983, 1987, 1991, 1993 and 1995). The American years were those in which the NJRP (National Judicial Reporting Program) survey was carried out (1986, 1988, 1990, 1992 and 1994), which yielded the number of adults convicted and sentenced to custody, plus 1981 and 1983, when estimates of these quantities were derived. The English years were those in which the BCS was carried out, which estimated the number of crimes committed and the number reported to the police. All of the needed information was not available for other years. The fourth publication aimed to present key results in a more user- friendly way, using graphics rather than tables. Also, the estimation methods used in the fourth publication were greatly simplified compared with the first three publications. This report is modeled on the fourth publication and aims to extend all the analyses for eight countries to study time trends between 1981 and 1999. The template for each chapter Each author was asked to follow the same organization and address the same topics in each chapter: 1. Provide a brief description of the country and its criminal justice system, including an impressionistic account of developments in criminal and penal policy and other relevant changes in the society between 1981 and 1999. 2. Provide brief definitions of the six crimes. Residential burglary includes attempts. Vehicle theft (including taking and driving away) includes thefts of mopeds and motorcycles; generally, attempts are excluded from victim surveys but included in police figures. Robbery includes attempts. Only serious assault is counted; generally, attempts are excluded from victim surveys but included in police figures. Rape (of females by males) is measured only in police data; attempts are included. Where figures are given only for serious sex assault, an estimate is made for rape. Homicide includes murder, manslaughter and infanticide; attempts are excluded. As far as possible, one victim equals one crime; where figures are given for incidents rather than victims (for example of robbery), the number of victims is estimated. Authors were asked to discuss changes over time (since 1980) in laws that affected the six offenses, and to specify adjustments made to maximize comparability over time. Also, they were asked to specify changes in the quality of crimes over time (such as the percentage of robberies involving firearms) where possible. 3. Specify sources of data. Authors were asked to specify the sources of all numbers so that the data could (in principle) be replicated in the future. 4. Describe victim survey data. Authors were asked to describe the victim surveys briefly, including design, sample sizes, response rates, sampling frame and coverage, method (that is face-to-face versus telephone?) and measures to combat telescoping. Specialized victim surveys were distinguished from omnibus social surveys including victimization questions. Authors were asked to provide raw data (numbers) plus confidence intervals where possible. Residential burglary and vehicle theft rates were specified per household, and robbery and serious assault rates were specified per population covered in the survey (for example all those age 16 or older). Victim survey data on rape (and, of course, homicide) were not presented. Differences over time (since 1980) in data collection procedures that affect the four victim survey crimes were specified, as well as adjustments that were made to maximize comparability over time. 5. Describe police data. Authors were asked to define police-recorded crimes, and to specify the precise step in processing that the crime was classified (such as when it is first reported?). Changes over time (since 1980) in police recording procedures that affected the six crimes were specified, as well as adjustments that were made to maximize comparability over time. Authors were asked to provide raw data (numbers) for each year and rates per population at risk. 6. Estimate the probability of police recording a reported offense. In order to estimate this, it was necessary to estimate the number of police-recorded crimes that were comparable to victim survey crimes. For burglary comparable police-recorded crimes are residential burglaries. For vehicle theft comparable police-recorded crimes are completed (not attempted) thefts of non-commercial vehicles. For robbery comparable police-recorded crimes are non-commercial robberies of victims over the minimum age for the victim survey. For serious assault comparable police-recorded crimes exclude victims under the minimum age for the victim survey. Then: B = R/D (1) Where B = Probability of the police recording a reported crime R = Number of comparable crimes recorded by the police D = Number of crimes reported to the police according to the victim survey. 7. Estimate the average number of offenders per crime. Authors were asked to obtain the best possible estimate of this quantity, from victim surveys, police records or self-reported offending data. If necessary subnational data or special surveys was used. Where the estimate was robust (such as from police records), the raw data for each year was used. Where the estimate was less robust (such as victim survey reports from burglary victims, where known offenders may be a small unrepresentative fraction of all cases), this quantity was averaged over all years. 8. Specify the number of convictions (that is persons convicted). Authors were asked to define the meaning of a "conviction", together with the minimum ages for juvenile and adult court. Changes over time (since 1980) that affected the six crimes were specified, as well as the adjustments that were made to maximize comparability over time. Raw data (numbers) were presented each year and rates per population at risk (for example those over the minimum age for conviction). Problems created by foreigners who may be in the conviction numerator but not in the population denominator were discussed. 9. Estimate the number of convictions per 1,000 offenders. This was calculated using the following equations: N = V*O (2) Where N = Number of offenders who could in principle have been convicted (based on victim survey crimes) V = Number of victim survey crimes O = Average offenders per crime Where victim survey and police-recorded crimes are not comparable, it is necessary to scale up from victim survey crimes to police- recorded crimes: M = N*P/R (3) Where M = Number of offenders who could in principle have been convicted (based on police-recorded crimes) P = Number of police-recorded crimes R = Number of police-recorded crimes that are comparable to victim survey crimes The number of convictions per 1,000 offenders is estimated as follows: X = C*1,000/M (4) Where X = Number of convictions per 1,000 offenders C = Number of persons convicted. Of course, X is not the number of convictions per 1,000 different offenders, but is the number of convictions (occasions that a person is convicted) per 1,000 times that an offender commits a crime. 10. Specify the number of custodial sentences. Authors were asked to describe definitions of custody and different types of custodial sentences available for juveniles and adults. Secure hospital orders were included as custodial sentences, but suspended sentences were not. Changes over time (since 1980) were specified that affected the six crimes, as well as adjustments that were made to maximize comparability over time. The raw number of custodial sentences each year, and rates per population at risk, were given. The probability of custody following a conviction was calculated each year. The probability of custody per offender was calculated as follows: Y = X*S (5) Where Y = Number of custodial sentences per 1,000 offenders S = Probability of custody following a conviction 11. Specify average sentence length and average time served. Authors were asked to calculate the average sentence length and average time served per custodial sentence, for each category of crime in each year. The basis of the estimates were described. Changes over time (since 1980) that affected the six crimes were specified, as well as adjustments that were made to maximize comparability over time. For life sentences for homicide, the effective sentence length was estimated using the following equation: L = T/F (6) Where L = Effective sentence length T = Average time served for homicide on a life sentence F = Fraction of non-life sentences for homicide that are served in custody. The average time served per offender was calculated using the following equation: Z = Y*D/1,000 (7) Where Z = Average time served per offender D = Average time served per custodial sentence Months were converted into days by multiplying by 30.44 (365.25/12). Authors were asked to produce spreadsheets and graphs that were comparable to those in the chapter on England. Even if national victim survey data were available only for a limited number of years, authors were asked to present national police, conviction and custody data for all available years in the graphs. Linking probabilities (for example the number of convictions per 1,000 offenders) could only be calculated for years where there was victim survey data available, and so only these years were shown in spread- sheets. 12. Summarize time trends. Authors were asked to summarize trends over time in the following key measures: (a) crime rates (according to victim surveys and police records). (b) probability of reporting given a crime, probability of recording given a reported crime. (c) conviction rates per population, and convictions per 1,000 offenders. (d) number of custodial sentences per population, and number of custodial sentences per 1,000 offenders. (e) average sentence length, average time served, fraction of sentence served in custody. (f) time served per 1,000 offenders. Where time trends were reasonably linear, authors were asked to calculate correlations between key indicators (such as crime rates) and the year. It was considered that correlations would provide some indication of the magnitude of time trends. Also, authors were asked to present correlations among all survey and recorded crimes. 13. Discuss possible explanations of time trends. In discussing explanations of time trends, authors were asked to calculate correlations between crime rates (survey and recorded) and key national indicators such as demographic factors, unemployment rates, measures of prosperity and income disparity, and criminal justice measures such as the probability and severity of punishment, the number of police officers and the financial costs of police, courts and prisons (corrected for inflation);. It was realized that it would not be possible to infer causal relationships from these correlations and that multivariate analyses were needed. However, it was considered that the basic (and time-lagged) correlations would provide useful information. Authors were asked to discuss issues that they considered to be important in explaining time trends, such as crimes by foreigners and drug addicts, and the percentage of persons with insurance. They were also asked to specify what future research was needed to test key hypotheses and what improvements are needed in national criminal justice data systems. Issues of comparability Our authors made Herculean efforts to comply with the template and to achieve comparability over time. Their efforts to comply with the template were facilitated by two meetings in Cambridge in which there were detailed discussions about crime and criminal justice in all eight countries, leading in some cases to changes in the original template. (The final template is listed above.) However, as described fully in the eight chapters, there are still problems of comparability. The most important of these concern crime definitions, victim surveys, and time served. There were fewer problems of comparability in regard to police-recorded crimes, persons convicted, persons sentenced to custody, and sentence length. In regard to crime definitions and legal codes, there was a major problem in distinguishing between serious and minor assaults, and serious assaults in one country are not very comparable to serious assaults in another. Changes over time in rates of serious assaults are more valid than comparisons between countries. Burglary and vehicle theft caused difficulties in Continental European countries (Netherlands, Sweden, Switzerland) because they were not distinguished explicitly from other types of theft in legal codes. However, the authors were able to estimate the numbers of these crimes through various adjustments. Between 1981 and 1999 the laws on rape in several countries were changed to define a more comprehensive crime (including anal and oral sex, including males as victims and females as offenders, and including acts between husbands and wives). Authors made various adjustments to estimate the number of rapes with male offenders and female victims in an effort to make the numbers comparable over time and between countries. During the same time period, the wording of questions in the national victim survey in several countries was changed in order to reveal more domestic violence, but again authors made adjustments to make the numbers comparable over time. Robbery and homicide were more comparable crimes. The United States was the only country which had a large-scale national victim survey every year. Sweden had a large-scale omnibus survey every year containing some victimization questions, but unfortunately it did not provide data on robbery. The Netherlands had three different national victim surveys that permitted annual estimates, but their results were not totally concordant. England had eight large-scale national victim surveys between 1981 and 1999, Scotland had five, Switzerland had five, Canada had three (plus a large-scale city survey which permitted national estimates), and Australia had three (but national estimates could also be derived from annual surveys in New South Wales, which accounted for about two-thirds of Australia's crime). Clearly, conclusions about trends and correlations based 5 or fewer years are fragile. The average time served was estimated in different ways in different countries. In Switzerland it was available in a sophisticated correctional database. In the Netherlands offenders serve a fixed proportion of their sentences. In England the estimate of average time served was based on release cohorts of prisoners. In the United States the fraction of time served (based on release cohorts) was applied to sentences given to estimate the time expected to be served. In Scotland and Sweden, the expected time to be served was estimated from laws and parole regulations. In Australia the average time served was estimated from the expected time to be served by the population of prisoners (obtained in a prison census); unfortunately, the daily population contains relatively more long-serving prisoners than entering or release cohorts. In Canada it was not possible to derive a satisfactory estimate of time served. Our attempts to obtain comparable data in eight countries highlights the problems and inadequacies of existing crime and criminal justice data. To be fair, the quality of data improved between 1981 and 1999. However, in no country was there a satisfactory measure of the number of offenders per crime, which is essential for linking crime data with offender data. An obvious recommendation is that, whenever the police record a crime in any country, they should record (as far as is known) how many persons committed the crime. Also, whenever a person is convicted in any country, the records should indicate how many other persons committed each crime (and also their identities, so that co-offenders can be linked). Cross-national comparisons Many cross-national comparisons can be carried out using the data presented in the following eight chapters. As an illustration, we focus on the property crime of burglary and on the violent crime of robbery. We focus on burglary and robbery because (a) survey crime data are not available for rape and homicide, (b) problems of comparability between countries are greatest for assault, and (c) we judged motor vehicle theft to be a less interesting property crime than burglary. We concentrate on 2 measures of crime (the survey crime rate and the recorded crime rate) and on 3 measures of punishment (the number of convictions per 1,000 offenders, the probability of custody following a conviction, and the average time served in custody) because we judged these as the most important. Burglary We show changes in the survey crime rate for all eight countries. The survey burglary rate was consistently highest in Australia and lowest in Switzerland (followed by Sweden). It increased in England up to 1993 and then decreased, while it decreased steadily in the United States. There were correlations between the survey burglary rate and the year. Correlations were positive in England (.67) and Switzerland (.76), negative in the United States (-.97) and Canada (-.83), and low in the other four countries. It should be borne in mind that these correlations are based on only four years for Canada and five years for Scotland and Switzerland. Changes in the recorded burglary rate for all eight countries varied. This was highest in Scotland and Australia and lowest in Switzerland and Sweden. It increased steadily in Australia and the Netherlands and decreased steadily in the United States. Correlations between the recorded burglary rate and the year were positive for Australia (.97), Switzerland (.87), the Netherlands (.78), and England (.48) and negative for the United States (-.95), Scotland (-.64), Canada (-.56), and Sweden (-.44). Changes in the number of convictions per 1,000 burglary offenders is a measure of the probability of punishment. Canada is excluded from this figure because conviction data were not available before 1994. The number of convictions per 1,000 offenders was highest for Australia and Scotland and lowest for Sweden or England. The correlations show that it decreased over time in England (-.89), Sweden (-.84), Scotland (-.83), Netherlands (-.87), Australia (-.84), and Switzerland (-.86), but increased in the United States (.82). The probability of custody following a conviction for burglary is a measure of the severity of punishment (figure 1d). This was highest in the United States and Sweden (at least until 1994) and lowest in Australia. The correlations show that this probability increased over time in Scotland (.93), England (.60), Australia (.53), the United States (.41), and Switzerland (.40), but decreased over time in Sweden (-.63). The average time served for burglary, is a measure of the severity of punishment. This was highest in the United States (until 1994) and lowest in Sweden and Scotland. The correlations show that this increased over time in Australia (.82), England (.74), Switzerland (.62), Sweden (.53), and Scotland (.43) but not in the United States (-.12). The survey and recorded crime rates for burglary were correlated over time in the United States (.98), Switzerland (.96), England (.91), Canada (.76), Scotland (.74), and the Netherlands (.49), but less so in Sweden (.26) and not at all in Australia (-.10). The number of convictions per 1,000 offenders was negatively correlated with the survey crime rate in all countries, but most strongly in England (-.89), the United States (-.86), and Switzerland (-.85) and hardly at all in the Netherlands (-.10). Correlations between the probability of custody following a conviction and the survey crime rate were substantial and negative only for the United States (-.48) and Scotland (-.43). There were no substantial negative correlations between the average time served and the survey crime rate, and one substantial positive correlation for Australia (.40). Robbery The survey robbery rate was highest in Canada and the Netherlands, and lowest in Scotland (until 1995). In 1999 the survey robbery rate was lowest in the United States. It was not available for Sweden. The correlations show that the survey robbery rate increased over time in England (.91) and Scotland (.74) but decreased over time in the United States (-.69). Correlations in the other four countries were low. The recorded robbery rate shows this was highest in the United States (until 1998) and lowest in Switzerland (figure 2b). The correlations show that it increased in Australia (.97), England (.96), the Netherlands (.93), Sweden (.91), Switzerland (.80) and Scotland (.49), decreased in the United States (-.43) and did not change in Canada (.07). The probability of custody following a conviction for robbery was lowest in Switzerland and usually highest in Sweden and the United States (figure 2d). The correlations show that it decreased over time in Switzerland (-.92), Australia (-.90), Sweden (-.73), and England (-.60) and increased in Scotland (.81). The average time served for robbery was highest in the United States and Australia and lowest in the Netherlands. According to the correlations, it increased over time in Switzerland (.99), England (.92), and the Netherlands (.79) but did not change markedly in the other four countries. The survey and recorded crime rates were highly correlated in England (.94) and the United States (.81) but not in any other country. The negative correlation in Canada (-.81) was based on only 4 years. The number of convictions per 1,000 offenders was negatively correlated with the survey robbery rate in Switzerland (-.93), Scotland (-.83), the Netherlands (-.71), and the United States (-.66), but the correlations were low in England (-.10) and Australia (-.17). The probability of custody following a conviction for robbery was negatively correlated with the survey robbery rate only in England (-.56); the correlation was positive in Scotland (.55). The average time served for robbery was negatively correlated with the survey robbery rate only in the Netherlands (-.63); the correlation was positive in England (.79). The number of convictions per 1,000 robbery offenders was generally high in Scotland and the United States and generally low in Switzerland and England. The correlations show that it increased over time in the Netherlands (.77), the United States (.74), and Australia (.45), and decreased over time in England (-.88), Switzerland (-.62), and Scotland (-.33). References Farrington, D. P. and P. A. Langan, (1992) "Changes in crime and punishment in England and America in the 1980s." Justice Quarterly, 9, 5-46. Farrington, D. P., P. A. Langan, and P-O. H. Wikstr"m. (1994) "Changes in crime and punishment in America, England and Sweden between the 1980s and the 1990s." Studies on Crime and Crime Prevention, 3, 104-131. Farrington, D. P. and P-O. H. Wikstr"m, (1993) "Changes in crime and punishment in England and Sweden in the 1980s." Studies on Crime and Crime Prevention, 2, 142-170. Federal Bureau of Investigation, (2002) Crime in the United States, 2001. Washington, DC: U.S. Department of Justice. Kesteren, J. van, P. Mayhew, and P. Nieuwbeerta, (2000) Criminal Victimization in 17 Industrialized Countries. The Hague, Netherlands: Ministry of Justice. Langan, P. A. and D. P. Farrington, (1998) Crime and Justice in the United States and in England and Wales, 1981-96. Washington, D.C.: Bureau of Justice Statistics. Authors David P. Farrington, O.B.E. (Officer of the British Empire), is Professor of Psychological Criminology at the Institute of Criminology, Cambridge University. He received B.A., M.A. and Ph.D. degrees in psychology from Cambridge University, the Sellin-Glueck Award of the American Society of Criminology for international contributions to criminology, and the Sutherland Award of the American Society of Criminology for outstanding contributions to criminology. His major research interest is in the longitudinal survey of delinquency and crime, and he is Director of the Cambridge Study in Delinquent Development, a prospective longitudinal survey of over 400 London males ages 8 to 48. He is also co-investigator of the Pittsburgh Youth Study, a prospective longitudinal study of over 1,500 Pittsburgh males ages 7 to 25. Besides about 350 published papers on criminological and psychological topics, he has published 24 books, one of which (Understanding and Controlling Crime, 1986) won the prize for distinguished scholarship of the American Sociological Association Criminology Section. He is chair of the Campbell Collaboration Crime and Justice Group, a member of the board of directors of the International Society of Criminology, a member of the advisory board of the U.S. National Juvenile Court Data Archive, joint editor of the Cambridge Criminology Series and of the journal, Criminal Behaviour and Mental Health, and a member of 14 other journal editorial boards. He was the first person from outside North America elected president of the American Society of Criminology, and was president of the European Association of Psychology and Law, president of the British Society of Criminology, president of the Academy of Experimental Criminology, and an officer or member of other associations in the U.K., U.S., and Netherlands. Patrick A. Langan is Senior Statistician at the Bureau of Justice Statistics in the United States Department of Justice. He received a B.A. degree in sociology and M.A. and Ph.D. degrees in criminology from the University of Maryland. Formerly he was social science analyst at the National Institute of Justice, research statistician at the Maryland Department of Juvenile Services, and Research Fellow at the Institute of Criminal Justice and Criminology, University of Maryland. Michael Tonry joined the Institute of Criminology, Cambridge University as Director in October 1999. Formerly, he was Sonosky Professor of Law and Public Policy at the University of Minnesota Law School, a visiting fellow at All Souls College, Oxford University, and a senior fellow in the Meijers Institute of the faculty of law, University of Leiden. He also practiced law in large commercial law firms in Chicago and Philadelphia and as a sole practitioner in rural Maine. He was editor and publisher of The Castine Patriot, a weekly newspaper in Castine, Maine, and is author or editor of a number of books, including Thinking about Crime: Sense and Sensibility in American Penal Culture (2004), The Future of Imprisonment (2004), Sentencing and Sanctions in Western Countries, with Richard Frase (2001), The Handbook of Crime and Punishment (1998), Sentencing Matters (1996), and Malign Neglect: Race, Crime and Punishment in America (1995). He has worked as an advisor to federal and state agencies in the U.S., Australia, and Canada, to national government agencies in Europe, and to international organizations. He is editor of Crime and Justice: A Review of Research, published since 1979 by the University of Chicago Press, Overcrowded Times, a bimonthly sentencing and corrections newsletter established in 1989, and the book series, Studies in Crime and Public Policy, established in 1992 by Oxford University Press. Darrick Jolliffe is a researcher associated with the Institute of Criminology, University of Cambridge and the Forensic Psychiatry Research Department, Queen Mary and Westfield College, University of London. His research interests include individual factors and offending, criminal career research and the evaluation of offender treatment programs. His publications include: Empathy and Offending: a Systematic Review and Meta-analysis, Predictive, Concurrent, Prospective and Retrospective Validity of Self-Reported Delinquency, Evaluation of Two Intensive Regimes for Young Offenders and A Feasibility Study into Using a Randomised Controlled Trial to Evaluate Treatment Pilots at HMP Whitemoor. Conclusions The information in this report represents the most comprehensive and sophisticated effort to present comparable data on national crime rates and on the probability and severity of legal punishment. We hope that it will be useful in improving the ability to explain national trends in crime and punishment over time and differences between countries. England and Wales Aims The main aim of this chapter is to summarize trends in crime and justice in England and Wales between 1981 and 1999 and to investigate some possible explanations of them. Six serious offenses are studied: residential burglary, motor vehicle theft, robbery, serious assault, rape, and homicide. A criminal justice system involves a successive funneling process, shown below in a simplified form: Crime Committed Reported to Police Recorded by Police Offender Convicted Sentenced to Custody Sentence Length Time Served Prison Population Of all crimes committed only some are reported to the police. Of all crimes reported only some are recorded by the police. Of all crimes recorded only some lead to the detection of an offender and to a conviction in court. Of all offenders found guilty in court, only some are sentenced to custody. These offenders receive sentences of different lengths but they serve only a portion of their sentence in custody, thereby becoming the prison population. There are many other possible stages that could have been shown (for example arrest), and many disposals other than imprisonment, but these are some of the most important stages. The key questions addressed are: * Is the serious crime rate per capita increasing or decreasing? * Is the conviction rate per capita increasing or decreasing? * Is the probability of an offender getting convicted increasing or decreasing? * Is the probability of a convicted offender being sent to custody increasing or decreasing? * Is the average sentence length increasing or decreasing? * Is the average time served increasing or decreasing? The crime rate can be measured either from victimization survey data or from crimes recorded by the police. Generally, survey data are considered to provide the more accurate measure of crime. The probability of a victim reporting a crime to the police and the probability of the police recording a crime that is reported to them provide the links between victimization and police-recorded data, and help to understand both. Consequently, two further questions will be addressed: * Is the probability of a victim reporting a crime to the police increasing or decreasing? * Is the probability of the police recording a crime that is reported to them increasing or decreasing? This is the final important question: * Is the average time served per offender increasing or decreasing? The average time served per offender is a summary measure of punitiveness, combining the probability of an offender getting convicted, the probability of an offender receiving custody following a conviction, and the average time served. England and Wales Description Great Britain (GB) is the island containing England, Wales and Scotland, while the United Kingdom (U.K.) contains GB and Northern Ireland. Many statistics are collected for GB or the UK rather than for England and Wales (see Office for National Statistics or ONS, 2000a, 2001). In 1999 the resident population of England was 49.8 million, compared with 2.9 million in Wales. The population density of England is high (381 people per square kilometer, compared with 141 in Wales). Between 1981 and 1999 the population increased by 6.3% in England and by 4.4% in Wales. The UK is a Western industrialized democracy and a constitutional monarchy. Its members of parliament (MP's) are from England (529), Wales (40), Scotland (73), and Northern Ireland (18). Unlike England the other three countries have separate national bodies responsible for their central administration. The UK has an aging population, and projections are that the median age will continue to rise from its 1999 value of 37. Life expectancy is 75 for males and 80 for females. There are over 300,000 marriages and over 150,000 divorces each year in the UK, and there has been a big increase in single-person and single-parent households in recent years. The GB resident population contains 6.7% from non-white ethnic minority groups, of which the most prevalent are South Asians (Indians, Pakistanis, Bangladeshis -- 3.4%) and Blacks (Africans, Caribbeans -- 2.1%). The U.K. National Health Service provides a full range of medical services available to all residents regardless of income. All children between age 5 and 16 are required by law to receive full-time education, and about two-thirds of children age 3 and 4 attend preschool education. About one-third of people continue in higher education after age 18, and this has increased considerably in recent years. The United Kingdom is a relatively prosperous country. The value of all goods and services produced in the U.K. economy for final consumption is measured by the Gross Domestic Product (GDP). In 1999 the GDP at current market prices totaled £891 billion, or about £15,000 per person. The average annual growth in GDP at constant market prices between 1995 and 1999 was 2.9%. Inflation is currently low at about 2%. About 85% of males and 73% of females of working age are economically active, and the unemployment rate is historically low at 5.2% for England and 6.1% for Wales. In the United Kingdom over 70% of households have at least one car and over 25% have at least two. The criminal justice system England and Wales have a common criminal justice system, while Scotland and Northern Ireland have different systems. In England and Wales the Home Office is responsible for the criminal law, the police, prison and probation services, and it publishes most criminal justice statistics (Chapman and Niven, 2000). The Lord Chancellor's Department is responsible for the courts and the Attorney General is responsible for the Crown Prosecution Service. There is an adversarial system of justice and most cases are sentenced by a panel of three unpaid lay people in Magistrates' Courts. The more serious (indictable) offenses are processed by judges and juries in the Crown Courts. There are 43 police forces in England and Wales and a total of about 125,000 police officers, or about 1 officer per 420 persons. The minimum age of criminal responsibility in England and Wales is 10. Persons age 10 to 17 are tried separately from adults in special Youth Courts, although they may be tried in adult Magistrates' Courts or Crown Courts depending on the seriousness of their offenses. In 1999 the main custodial sentence for those age 15 to 20 was detention in a Young Offender Institution, while persons age 12 to 14 could be given Secure Training Orders. From April 1, 2000 the main custodial sentence for those age 12 to 17 became the Detention and Training Order. Persons age 21 or older can be sent to prison. Those receiving fixed sentences of less than 4 years are released after serving half the sentence, while those serving sentences of 4 years or more can be released by the Parole Board after serving half their sentence and must be released after serving two-thirds of their sentence (unless they misbehave in prison). After the Labour Government came to power in 1997, it pursued a two- pronged approach of increasing prevention and punishment. The Crime and Disorder Act 1998 especially focused on prevention, setting up Crime and Disorder partnerships in local areas to take charge of local crime reduction initiatives, and establishing Youth Offending Teams to oversee rehabilitative and restorative justice programs for young offenders. Regarding punishment the Crime (Sentences) Act 1997 specified that an offender age 18 or older on a second conviction for a serious violent or sexual offense should receive an automatic life sentence unless there are exceptional circumstances. It also specified that an offender age 18 or older on a third conviction for residential burglary should receive a minimum sentence of 3 years imprisonment, but this did not take effect until 2000. Method The criminal justice system in England and Wales can easily be described using a flow diagram such as that presented above. It is much more difficult to quantify the system in practice by specifying the exact number of offenders flowing through at each stage. Such a specification would have great theoretical and practical relevance. For example, it would help to determine whether changes in prison populations were caused by changes in crime rates, reporting, recording, conviction rates, the probability of custody, sentences given, or time served. All following references to England will include Wales (in order to simplify the exposition). A key problem is that national data tracking individual offenders by some unique identifier across the different stages of the criminal justice system are not available in England. Ideally a longitudinal study is needed, beginning with offenses and tracking offenders through the different stages of the system. However, aggregate national data are available for each of the stages separately (such as crimes committed, persons convicted, persons sentenced to custody). These separate counts do not arise from tracking the same individuals across stages, but they permit reasonably accurate estimates of the flow of offenders from one stage to the next. The present research This chapter is a continuation of previous work, which began with David Farrington's presidential address to the British Society of Criminology in January 1990, attempting to link national-level data in England to estimate offense-specific numbers flowing through the criminal justice system at each stage, from crimes committed to crimes reported to the police, crimes recorded by the police, offenders convicted, offenders sentenced to custody, average sentence length, and average time served, in two years (1981 and 1987). The first publication (Farrington and Langan, 1992) estimated numbers flowing through the criminal justice system in both England and the United States and compared trends over time (between 1981 and 1987 in England, between 1981 and 1986 in the United States). The first year that it was possible to make these estimates in England was 1981, which was the first year of the national victimization survey -- the British Crime Survey or BCS (Hough and Mayhew 1983). The BCS was repeated at 2-year or 4-year intervals, covering the years 1983, 1987, 1991, 1993, 1995, 1997 and 1999. Since 2001 (covering crimes in the year 2000), the BCS has been repeated annually. The sample size was initially just over 10,000, but it gradually rose to 19,400 for the year 1999, and it was projected to double to 40,000 in 2001 (for the year 2000). National figures for police-recorded crimes, convictions, and sentencing have been available annually in England for many years. There are no published national arrest figures for England. National figures for average sentence length and average time served were first published in 1986 (Home Office 1987, p.77). Offense-specific figures for average sentence length and average time served were provided for prison release cohorts in certain years by the Home Office. The second publication (Farrington and Wikström, 1993) estimated numbers flowing through the criminal justice system in England and Sweden and compared trends over time (between 1981 and 1987). The third publication (Farrington, Langan, and Wikström 1994) extended the previous analyses to between 1981 and 1991 for England and Sweden and between 1981 and 1990 for the United States. One problem with the first three analyses is that they reported changes between two widely separated time points rather than trends over time, making it difficult to know precisely when changes occurred and how they might be explained. The fourth publication (Langan and Farrington, 1998) calculated all numbers and probabilities for 7 years in the United States (1981, 1983, 1986, 1988, 1990, 1992 and 1994) and for 6 years in England (1981, 1983, 1987, 1991, 1993 and 1995). The American years were those in which the NJRP (National Judicial Reporting Program) survey was carried out (1986, 1988, 1990, 1992 and 1994), which yielded the number of adults convicted and sentenced to custody, plus 1981 and 1983, when estimates of these quantities were derived. The English years were those in which the BCS was carried out, which estimated the number of crimes committed and the number reported to the police. All of the needed information was not available for other years. The fourth publication aimed to present key results in a more user-friendly way, using graphics rather than tables. Also, the estimation methods used in the fourth publication were greatly simplified compared with the first three publications. This chapter uses the same methods as the fourth publication. Eight years are covered (1981, 1983, 1987, 1991, 1993, 1995, 1997, 1999). In general, sources of data are given only for the most recent years (1997 and 1999); for sources of data in earlier years, one should consult the earlier publications. While the key results are shown graphically, the full spreadsheets are included as tables 1-6. Readers who are mainly interested in substantive issues as opposed to methodology should now advance to the Results section. Comparability There are two main problems of comparability: over time and between victim survey and criminal justice data. Legal definitions of burglary (entering as a trespasser to commit theft or damage), robbery (theft involving force or threat), motor vehicle theft (theft or unauthorized taking of motor vehicles) and homicide (murder, manslaughter or infanticide) are relatively clear and constant over time. Attempts are included with completed crimes, except for vehicle theft in survey data and homicide in criminal justice data. Attempted vehicle theft was not included with vehicle theft in the BCS because of the difficulty of distinguishing between attempted theft of vehicles and attempted theft from vehicles. Attempted murder is a separate legal category in England. Between 1981 and 1999 the legal definitions of burglary, robbery, and homicide did not change in England. Unauthorized taking of a motor vehicle was downgraded from an indictable (more serious) to a summary (less serious) offense in the Criminal Justice Act 1988 (with effect from October 12, 1988). Subsequently a new offense of aggravated vehicle taking was created by the Aggravated Vehicle Taking Act 1991 (with effect from April 1, 1992). For comparability with earlier years, English motor vehicle theft in 1999 comprised four legal categories: theft of a motor vehicle (indictable), unauthorized taking of a motor vehicle (summary), and aggravated vehicle taking (indictable and summary). Rape (including attempted rape) is a more problematic offense. In England in 1981 rape had to involve a male offender and a female victim and required penetration of the vagina by the penis. Husbands could not be convicted of raping their wives. No male under age 14 could be convicted of rape. Female offenders were included in the rape statistics if they aided or abetted rape. The minimum age for a rape conviction was decreased from 14 to 10 -- the minimum age of conviction for other offenses -- in the Sexual Offenses (Amendment) Act 1993 (with effect from September 20, 1993). The definition of rape was changed in the Criminal Justice and Public Order Act 1994 (with effect from November 1, 1994) to include male victims, spouse victims, and anal intercourse. However, the majority of rape offenses continue to involve male offenders, female victims, and vaginal intercourse. Consensual sex with girls under 16 is placed in a different legal category. It is possible to identify the numbers of male offenders and female victims in the official criminal statistics. The sample size in the BCS is too small to yield a survey estimate of rape. Serious assault poses the greatest problem. In England up to 1997, the BCS (and the police-recorded figures) distinguished between the indictable offense of wounding (causing actual or grievous bodily harm) and the summary offense of common assault. Serious assault is defined as wounding in this chapter. Wounding occurs if the victim receives some kind of cut or wound, where the skin or a bone is broken, or if medical attention is needed. Common assault occurs if the victim is punched, kicked or jostled, with negligible or no injury. Minor bruising or a black eye count as negligible injury. Attempted assaults are not counted as wounding in the BCS. Because of legal problems such as establishing intention, most attempted assaults would also not be counted as wounding by the police. The legal definition of wounding did not change between 1981 and 1997. Ideally, it would be desirable to use an injury severity scale to classify the seriousness of assaults. The Home Office rules specifying how crimes are counted by the police did not change between January 1980 and March 1998. From April 1998 new counting rules were introduced (see Povey and Prime 1999). At the same time police-recorded crimes began to be presented in the annual Criminal Statistics on a financial year basis (for example, April 1998-March 1999) rather than on a calendar year basis, although data on convictions and sentencing continued to be presented for each calendar year. The new counting rules particularly affected the number of police-recorded criminal damage and fraud offenses. They had a negligible impact on five of our six offenses (all except serious assault). The main effect was on the "other wounding" category (8), which was extended to include possession of weapons (8B), harassment (8C) and (from April 1999), racially aggravated wounding (8D). However, the number of other woundings in the original category (8A) was still presented and Povey and Prime estimated that this number was only 3.2% less than the total number of other woundings under the old rules. Therefore, the number of police- recorded woundings in 1999 (categories 8A and 8D) may be about 3% less than in 1997 because of the change in the counting rules. Great efforts have been made to ensure that crime definitions in the BCS are comparable to those in official (police) statistics, but some differences are inevitable. For example, the BCS does not include thefts of commercial vehicles, crimes against organizations, or against persons under 16. However, one can adjust the police statistics to obtain figures which are comparable to victim survey estimates. Survey offenses The number of victim-survey offenses, comparable population figures (number of households and number of persons age 16 or older), and the probability of reporting to the police were obtained from the BCS (Budd, 2001; Kershaw and others 2000; Mirrlees-Black and others, 1998; Mattinson, 1999; 2000). The figures shown in tables 1-6 are the latest estimates for all years taking account of the 1991 Census. For example, the BCS estimated that there were 345,994 robberies in 1999, and that 30% of them were reported to the police. Since there were an estimated 41,996,000 persons age 16 or older in 1999, the survey robbery rate was 8.24 per 1,000 population at risk; disregarding repeat victims, about one in every 121 persons was robbed in 1999. All BCS figures, of course, have confidence intervals about them. For example, the 95% confidence interval for the robbery rate in 1999 was from 5.06 to 11.42 per 1,000 population. Confidence intervals are narrower for the other three offenses, which are more prevalent. BCS survey crime rates for residential burglary and vehicle theft are per 1,000 households, while rates for robbery and serious assault (wounding) are per 1,000 population age 16 or older. Vehicle theft figures refer to completed thefts only. Other population estimates came from the Office for National Statistics (such as 2000b); tables 1-6 show the latest estimates for all years. The main change in the BCS over the years was the addition of a new screening question for domestic violence ("Has any member of your household deliberately hit you with their fists or with a weapon of any sort or kicked you or used force or violence in any other way?") and a new victim form in 1993. This caused an increase in the number of victim-survey offenses of serious assault. Mayhew (1997) and Mattinson (1999, 2000) provided serious assault estimates with and without the new domestic violence screening question; for example, the figures were 810,994 (without) and 860,395 (with) in 1995. For comparability with the 1981-91 figures, the "without" figures are used in these analyses. In order to link up offenses and offenders, the average number of offenders per offense is needed. This is because one offense committed by two offenders can lead to two convictions (if both offenders are convicted). Thus, the number of offenders at risk of conviction is the number of offenses multiplied by the average number of offenders per offense. Merely comparing the number of offenses with the number of convictions is like comparing apples and oranges. For burglary, vehicle theft, robbery, and serious assault, the average number of offenders per offense was provided from the BCS by Mirrlees-Black (1996) and Mattinson (1999, 2000). For homicide this was obtained from the annual Criminal Statistics (Home Office, 1998a, p.71; 2000a, p.75). The average over all years was used in estimating probabilities. This was 1.8 for burglary, 2.1 for vehicle theft, 2.3 for robbery, 2.0 for serious assault, and 1.1 for homicide. For rape the only national data on the number of offenders per offense seems to be that published by Grace, Lloyd and Smith (1992) for offenses committed in 1985. They reported that, for police-recorded offenses, there were 1.1 offenders per rape offense. This figure was used in all calculations. Police-recorded offenses The number of police-recorded offenses was obtained from the annual criminal statistics. The number of recorded offenses in the calendar year 1999 was provided by Lavin (2001). The number of police-recorded offenses refers to the number initially recorded by the police in each offense category in each year, irrespective of later court proceedings. Assault comprised section 5 wounding (mainly causing grievous bodily harm) and section 8a and 8d wounding (mainly causing actual bodily harm). Vehicle theft comprised theft or unauthorized taking of motor vehicles and aggravated vehicle taking. Only rape of a female was counted. As an example, there were 78,884 recorded robberies in 1999. Since the resident population was estimated to be 52,690,000 in 1999, the police-recorded robbery rate was 1.5 robberies per 1,000 population (one robbery per 668 citizens). The numbers of police-recorded offenses that were comparable to survey offenses were estimated by Mayhew (1996), Mattinson (1999, 2000) and Budd (2001). The estimation procedures were explained by Mirrlees-Black and others (1998, pp. 75-83) and Kershaw and others (2000, pp. 115-121). In 1997 the total number of recorded robberies was 64,878 (63,072 recorded by the police and 1,806 recorded by the British Transport Police). Excluding an estimated 20% of cases where the victim was less than age 16, 51,902 of these recorded robberies were considered to be comparable to the 345,994 BCS robberies. The main adjustments to recorded crimes were to exclude thefts of commercial vehicles, attempted vehicle thefts, and victims under 16. However, robberies of business property were counted in the BCS. Dividing the number of comparable police-recorded offenses by the number of survey offenses yields the probability of a survey offense being recorded by the police. For robbery in 1999 this was .19 (64,472/345,994). It is more interesting to disaggregate this figure into the probability of a survey offense being reported and the probability of a reported offense being recorded by the police. Since the probability of reporting a robbery in 1999 was .30, the probability of a reported robbery being recorded by the police in 1999 was .61 (64,472/105,198). The fact that the estimated probability of a reported offense being recorded was greater than 1 for vehicle theft in 1981 and 1983 is probably a function of the confidence interval around the number of reported offenses and the fact that almost all of the reported offenses would have been recorded. Convictions The number of persons convicted for each offense was obtained from the annual Supplementary Criminal Statistics (for example, Home Office 1998b, Annex A; Home Office 2000b, Annex A). In earlier years it was necessary to add convictions in Crown Courts and Magistrates' Courts. It was also necessary to add different offense categories (murder, manslaughter, infanticide, and manslaughter due to diminished responsibility for homicide; section 5 and 8 woundings for assault; theft of a motor vehicle, unauthorized taking, and indictable and summary aggravated vehicle taking for vehicle theft). As already explained rape convictions since 1994 can include male and female victims and male and female offenders. The Supplementary Criminal Statistics showed that there were 601 male offenders with female victims in 1999. As an example, 5,626 persons were convicted for robbery in 1999. Since there were an estimated 46,029,000 persons age 10 or older in 1999 (Office for National Statistics, 2000b), the conviction rate for robbery was 0.122 per 1,000 population at risk. Since there were an estimated 345,994 robberies in 1999 and an estimated 2.3 offenders per offense, there were an estimated 795,786 offenders (not necessarily different persons) who could in theory have been convicted for BCS- comparable robberies if the criminal justice system had been 100% efficient. Since BCS-comparable robberies comprised 82% of all recorded robberies in 1999 (64,472/ 78,884), it could be estimated that there were 973,675 offenders at risk of conviction for all types of robberies. Dividing this number by the number of persons convicted (5,626) yields the estimate that there were 173 offenders per conviction, or that the probability of conviction for each robbery offender was .0058 in 1999 (5.8 convictions per 1,000 offenders). Alternatively, it might be said that the average robber could commit 173 robberies for every one court appearance leading to conviction. This neglects cautions (576 for robbery in 1999)and multiple offenses dealt with on one court appearance, on the assumption that one conviction provides one opportunity for legal punishment as far as the offender is concerned. The calculations for rape and homicide were based only on police- recorded offenses. While only one in 173 robbery offenders was convicted for each offense, the probability of a robbery offender being convicted sooner or later is much greater than this, because the average offender commits several offenses. For example, in the Cambridge Study in Delinquent Development, 58% of males who committed burglary between age 10 and 32 (according to self-reports) were convicted of burglary at some time (Farrington, 1989). National arrest data were collected for the first time in 1999 (Barclay, 2001). For example, there were 24,351 arrests for robbery. It can therefore be calculated that the probability of arrest for robbery in 1999 was .025 (24,351/ 973,675), or conversely that 1 in 40 robbery offenders was arrested. (This probability is an overestimate, because it is possible for a person to be arrested more than once for the same offense.) The probability of an arrest for robbery leading to a prosecution for robbery was .424 (10,321/24,351). The probability of a prosecution for robbery leading to a conviction for robbery was .545 (5,626/10,321). The multiplication of .025 (probability of arrest per offender) by .424(probability of prosecution per arrest) and by .545 (probability of conviction per prosecution) yields the figure of .0058 (probability of conviction per offender). Assuming that these data are collected in future years, it should be possible to estimate these system probabilities each year and investigate their trends over time. Probability of custody The number of persons sentenced to custody for each offense was obtained from the Supplementary Criminal Statistics (for example Home Office 1998b, Annex A; Home Office 2000b, Annex A). As before it was necessary to add Crown Courts and Magistrates' Courts and different offense categories. It was also necessary to add different types of custodial sentences (in 1999: imprisonment, detention in a young offender institution, secure hospital order, secure training order, detention under section 53 of the Children and Young Persons Act 1933; in 1981: imprisonment, borstal, detention center, secure hospital order, detention under section 53 of the Children and Young Persons Act 1933). As an example, 4,085 offenders were given custodial sentences for robbery in 1999, or 73% of all convicted robbery offenders. The incarceration rate for robbery in 1999 was 0.09 per 1,000 persons age 10 or older. The probability of an offender receiving a custodial sentence was calculated by multiplying the probability of an offender being convicted by the probability of a conviction being followed by custody. For robbery in 1999 this was .0042, since there were 5.8 convictions per 1,000 robberies and 73% of convictions for robbery were followed by a custodial sentence. This corresponds to one in every 238 robbers receiving a custodial sentence (4.2 incarcerations per 1,000 offenders). Again, these are not necessarily 238 different persons. To the extent that people commit several robberies, their probability of receiving a custodial sentence sooner or later will be much greater than this. Sentence length and time served Offense-specific data on average sentence length and average time served in prison by released prisoners (including juveniles) in 1997 and 1999 were supplied by Stevens (1999, 2000). The figures were provided for burglary in general, not for residential burglary specifically. Figures for offenders released in 1982, 1984, 1988, 1991, 1993, 1995, 1997 and 1999 are shown in tables 1-6. Initially it was thought that the following year was the most relevant (for example, because most robbers sentenced to custody in 1981 were released in 1982). However 1992 data were not available at the time of the analysis by Farrington, Langan, and Wikström (1994), so 1991 data were used, and for consistency 1993, 1995, 1997 and 1999 data were used subsequently. Offense-specific data on average sentence lengths were published for the first time in the 1995 Supplementary Criminal Statistics, but these figures were not used in this report (in the interests of comparability with earlier years). For example, these figures show an average sentence length for robbery of 35.3 months in 1999, compared with 40 months. The discrepancy may be attributable to short-sentence prisoners who are released immediately after their convictions (because of time served on remand) and hence are never received into prison after conviction. These would not be included in the figures for released prisoners. Alternatively, it may reflect a difference between those sentenced in 1999 and those released in 1999. Those released in 1999 are arguably more comparable to those sentenced in 1997, and the average sentence length for robbery in 1997 was 39.4 months, very close to the average sentence length of offenders released in 1999. It is quite complex to estimate average sentence length from the Supplementary Criminal Statistics, because averages are given separately for each type of sentence, for the Crown Court and Magistrates' Court, and for males and females. Thus, the 4,085 persons sentenced to custody for robbery in 1999 had the following average sentence lengths (excluding 37 who received life imprisonment): Crown Court prison 1960 males (45 months), Crown Court prison 119 females (26 months), Crown Court YOI (Young Offender Institution) 1,439 males (28 months), Crown Court YOI 81 females (20 months), Crown Court Section 53 detention 245 males (36 months), Crown Court Section 53 detention 12 females (27 months), Crown Court Secure Training Order (STO) 4 males (9 months), Magistrates' Court YOI 171 males (3.6 months), Magistrates' Court YOI 7 females (3.8 months), Magistrates' Court STO 9 males (4.7 months), Magistrates Court STO 1 female (6 months). The total sentence length was 143,077 months for 4,048 persons, or 35.3 months on average. The offense categories of prisoners are slightly different from those used in the Criminal Statistics, but "assault" and "wounding" in the prison data are approximately equivalent to the two "wounding" categories (5 and 8) in the Criminal Statistics, and "taking and driving away" in the prison data covers at least 95% of those sentenced to custody for vehicle theft in the Criminal Statistics (Barclay, 1993). The average sentence length and average time served for assault were weighted combinations of the assault and wounding categories (weighted by the number of released offenders). As an example, the average sentence length for robbery releasees in 1999 was 40 months, and the average time served was 23 months, or 58% of the sentence. The average time served per conviction was estimated by multiplying the probability of custody following a conviction by the average time served. For robbery in 1999 73% of convictions were followed by custody, and the average time served was 23 months (700 days), yielding an average of 508 days served per conviction. The average time served per offender is a summary measure of overall punitiveness that combines the probability of an offender being convicted (a measure of the risk of punishment) and the average time served per conviction (a measure of the severity of punishment). For robbery in 1999, one in 173 offenders was convicted, and 508 days were served per conviction for robbery, so an average 2.94 days were served per robbery offender. There are no English national data routinely published on time served in custody before conviction. However, the report of the Carlisle committee (1988, p.147) based on 1987 releasees estimated that about 10% of a prisoner's sentence was spent on remand before sentencing. The figures on time served after conviction in the Tables have not been adjusted to take this time into account. Homicide Special estimation procedures were used for homicide, because of the large number of indeterminate life sentences. Few life sentences were given for the other offenses. In England a life sentence (or an equivalent indeterminate detention for a juvenile) is mandatory for murder. In 1999 228 life sentences were given for murder, 16 life sentences were given for manslaughter, and 173 non-life prison sentences were given for manslaughter (Home Office 2000b, Tables S2.4, S2.5 and Annex A). There were also 32 section 53 orders and 25 secure hospital orders for homicide. Apart from these 59% (244/417) of the custodial sentences for homicide in 1999 were life sentences. In 1999 the average time served by homicide offenders first released from life sentences was 13.3 years or 159.6 months (Stevens, 2000). Of course the concept of sentence length is problematic with life sentences. The effective length of life sentences was estimated on the basis of the fraction of time served by homicide cases with non-life sentences. For non-life homicide offenders released in 1999, the average sentence length was 46.9 months and the average time served was 27.3 months, or 58% of the sentence (Stevens, 2000). Assuming that life sentence homicide cases were also serving 58% of their sentence, the effective length of a life sentence for homicide in 1999 was considered to be 274.2 months (159.6/.582) or 22.9 years. The average sentence length and average time served for homicide were weighted averages of life and non-life sentences. For example, in 1999 the average sentence length was 244 x 274.2 (life sentences) + 173 x 46.9 (non-life sentences), divided by 417, which came to 179.9 months. Results Survey crime rates Based on the national victimization survey (the BCS), the residential burglary rate per household more than doubled between 1981 and 1993 (from 41 to 86 per 1,000 households), then decreased by one-third up to 1999. The vehicle theft rate increased by two-thirds between 1981 and 1993 (from 16 to 26 per 1,000 households), but then decreased back to below the 1981 figure by 1999. The robbery rate almost doubled between 1981 and 1999 (from 4.2 to 8.2 per 1,000 population age 16 or older. The serious assault (wounding) rate increased by 50% between 1981 and 1995 (from 13.1 to 19.7 per 1,000 population age 16 or older), but then decreased by 30% between 1995 and 1999. In order to determine whether crimes were increasing markedly over time, the survey crime rate was correlated with the year. This correlation was based on only 8 years. Since the correlation coefficient can vary from 0 (no relationship) to 1 (a perfect relationship), a correlation of .5 or greater was considered to indicate a strong relationship. Burglary (r = .67), robbery (r = .91) and assault (r = .62) increased markedly between 1981 and 1999, whereas vehicle theft (r = .22) did not. The vehicle theft rate had a low correlation with year because of the nonlinear relationship: it increased up to 1993 and then decreased. The burglary rate varied similarly over time. Consequently, for burglary and vehicle theft, the correlation between crime rate and year is shown separately for two periods, 1981-93 and 1993-99. Both the burglary rate (r = .97) and the vehicle theft rate (r = .98) increased markedly between 1981 and 1993, and conversely both the burglary rate (r = -.94) and the vehicle theft rate (r = -.98) decreased markedly between 1993 and 1999. Recorded crime rates Like the survey burglary rate, the police-recorded residential burglary rate doubled between 1981 and 1993 (from 7.1 to 14.1 per 1,000 population), but then decreased by 39% up to 1999. Recorded crime rates are shown for all years from 1981 to 1999. Similarly, the vehicle theft rate increased by 73% between 1981 and 1993 (from 6.7 to 11.6 per 1,000 population)but then decreased back to 7.2 per 1,000 population in 1999. The robbery rate almost quadrupled between 1981 and 1999 (from 0.4 to 1.5 per 1,000 population;). The serious assault (wounding) rate more than doubled between 1981 and 1997 (from 2.0 to 4.6 per 1,000 population) but then decreased by 10% up to 1999. The police-recorded rape rate increased 7-fold between 1981 and 1999 (from 0.04 to 0.29 per 1,000 females;). The homicide (murder and manslaughter) rate increased by a quarter between 1981 and 1987 (from 0.011 to 0.014 per 1,000 population;), but then stayed tolerably constant. Based on 19 years for recorded crimes, the correlations show that recorded robbery (r = .96), assault (r = .97), rape (r = .99), and homicide (r = .81) all increased substantially over time in. Recorded burglary and vehicle theft increased substantially between 1981 and 1993 and decreased substantially between 1993 and 1999. Reporting crimes to the police According to victims the probability of reporting crimes to the police stayed tolerably constant over time for vehicle theft (mean 96%) and assault. The probability of reporting a burglary to the police increased up to 1991 (at 73%) but then decreased to 62% in 1999. The probability of reporting a robbery to the police increased to a peak in 1995-97 (57%) but then decreased remarkably, to 30% in 1999. All correlations between percent reported and year are low, confirming that the probability of reporting did not vary markedly over time. The large fluctuation in the probability of reporting a robbery may reflect estimates based on small numbers. In 1999 the BCS sample size was 19,411 persons age 16 or older out of a total population of 41,995,700. Therefore each survey robbery was scaled up by 2,164 (41,995,700/19,411) to produce a population estimate. Therefore the 345,994 survey robberies in 1999 represented 160 robberies (345,994/2, 164) disclosed to interviewers, of which 49 were reported to the police. The comparable estimates in 1997 were 112 robberies disclosed to interviewers and 62 reported to the police. These small numbers are likely to produce large year-to-year fluctuations in estimates. Recording crimes by the police The probability of the police recording a residential burglary that was reported to them decreased steadily from 70% in 1981 to 50% in 1997 but then increased to 59% in 1999. The probability of the police recording a vehicle theft also tended to decrease, but more irregularly, from 100% in 1981 to 83% in 1995 but it then increased to 91% in 1999. The probability of police recording decreased markedly over time for burglary (r = -.90) and vehicle theft (r = -.72). The probability of the police recording an assault increased irregularly, from 41% in 1981 to 67% in 1997 before decreasing to 54% in 1999. The probability of the police recording a robbery doubled from 1981 (24%) to 1991-93 (47%), then decreased back to 30% in 1997 before doubling again to 61% in 1999. Some of these large fluctuations obviously reflect estimates based on small numbers, as explained above. The probability of police recording increased markedly over time for robbery (r = .55) and assault (r = .79). Conviction rates The conviction rate for residential burglary decreased steadily between 1981 and 1999 (from 0.57 to 0.36 per 1,000 population age 10 or older;). Adding in cautions would not greatly alter this trend. The conviction rate for vehicle theft decreased steadily from 1981 to 1995 (from 0.83 to 0.34 per 1,000 population; and then remained constant. The conviction rate for robbery increased by a quarter between 1981 and 1999, while the conviction rate for assault decreased by half between 1981 and 1995 before increasing again. The conviction rate of males for rape increased by three-quarters between 1981 and 1999, while the conviction rate for homicide increased irregularly by a quarter. Conviction rates decreased markedly over time for burglary (r = -.92), vehicle theft (r = -.96) and assault (r = -.80) and increased markedly for robbery (r = .89), rape (r = .81) and homicide (r = .62). Custody rates The population custody rate (persons sentenced to custody per 1,000 population) for residential burglary decreased from 0.26 in 1983 to 0.16 in 1991 and then increased slightly to 0.22 in 1999. The custody rate for vehicle theft decreased dramatically from 0.21 in 1981 to 0.07 in 1991, but then increased somewhat to 0.10 in 1995-99. The custody rate for robbery stayed tolerably constant between 1981 and 1995 but then increased slightly. Similarly, the custody rate for assault stayed tolerably constant between 1981 and 1995 but then increased. The custody rates for rape and homicide closely tracked the conviction rates in increasing irregularly over time. The population custody rate increased markedly over time for robbery (r = .73), rape (r = .83), and homicide (r = .74), and decreased markedly for vehicle theft (r = -.70). Probability of an offender being convicted The number of convictions per 1,000 residential burglary offenders decreased steadily between 1981 and 1993, from 18.5 to 5.6 before increasing to 7.3 in 1999. The number of convictions per 1,000 vehicle thieves collapsed from 1981 to 1995 (from 51.6 to 11.7) but then increased to 16.5 in 1999. The number of convictions per 1,000 robbers stayed constant from 1981 to 1991 (at about 10) but then decreased to 5.8 in 1999. The number of convictions per 1,000 assaulters decreased from 1981 to 1995 (from 41.2 to 14.4) but then increased to 24.6 in 1999. The number of rape and homicide offenders was estimated from police-recorded crimes, not from victim survey crimes. The number of convictions per 1,000 rapists decreased from 272 in 1981 to 71 in 1999. The number of convictions per 1,000 homicide offenders fluctuated considerably around 600. The probability of an offender being convicted decreased substantially over time for burglary (r = -.89), vehicle theft (r = -.90), robbery (r = -.88), assault (r = -.87), and rape (r = -.95). This probability decreased between 1981 and 1993 for burglary (r = -.98) and vehicle theft (r = -.99), but then increased between 1993 and 1999 for burglary (r = .92) and vehicle theft (r = .92). Decreases in the probability of conviction in the 1980's could have been caused by (1) the increasing use of recorded cautions and unrecorded warnings for detected offenders (Home Office 1985, 1990b; Farrington, 1992); (2) the Police and Criminal Evidence Act 1984, implemented in January 1986, which increased procedural safeguards for accused persons (Irving and MacKenzie 1989); (3) the introduction of the Crown Prosecution Service in 1986, with lawyers replacing police officers as prosecutors, leading to an increasing tendency to drop cases rather than prosecute them, and (4) the requirement for advance disclosure of the prosecution case in May 1985. There were also measures affecting specific offenses. For example, new charging standards for assault were introduced on August 31, 1994, which downgraded some convictions from indictable wounding to nonindictable common assault. Probability of an offender receiving custody The probability of a residential burglary offender receiving a custodial sentence decreased between 1981 and 1993 (from 6.9 to 2.3 incarcerations per 1,000 burglars), but then increased to 4.5 in 1999. The probability of a vehicle thief receiving a custodial sentence decreased dramatically between 1981 and 1991 (from 13.0 to 2.5 incarcerations per 1,000 offenders), but then increased to 4.8 in 1999. The probability of a robber receiving a custodial sentence decreased irregularly from 1981 to 1999, from 7.4 to 4.2 incarcerations per 1,000 offenders. The probability of an assaulter receiving a custodial sentence decreased from 1983 to 1995 (from 7.4 to 3.9 incarcerations per 1,000 offenders), but then almost doubled up to 1999. The probability of a rapist receiving a custodial sentence decreased by three-quarters from 1981 to 1999, from 240 to 69 incarcerations per 1,000 offenders. The probability of a homicide offender receiving a custodial sentence showed no clear trend over time. The probability of custody per 1,000 offenders decreased over time markedly for burglary (-.75), vehicle theft (r = -.79), robbery (-.89), and rape (-.96). Probability of custody after a conviction The probability of receiving a custodial sentence after a conviction for residential burglary did not increase markedly between 1981 and 1993 (42%) but then increased to 62% in 1999. The probability of custody following a conviction for vehicle theft decreased from 25% in 1981 to 14% in 1991, but then increased to about 30% in 1995-99. The probability of custody after a conviction for robbery decreased between 1987 (79%) and 1995 (67%) but generally stayed around the 70% level. The probability of custody after a conviction for assault stayed tolerably constant from 1981 to 1993 at less than 20% but then increased to about 28% in 1995-99. The probability of custody after a conviction for rape was always very high. The probability of custody after a conviction for murder or manslaughter was also very high. The probability of custody after a conviction increased over time for burglary (r = .60), assault (r =.80), rape (r = .65) and homicide (r = .70)and decreased over time for robbery (r = -.60). For vehicle theft this probability decreased in 1981-93 (r = -.69) but increased in 1993-99 (r = .70). The decreases in the probability of custody in 1987-91 could have been caused by Home Office pronouncements encouraging judges and magistrates to avoid sending offenders to prison as far as possible, especially for nonviolent offenses such as burglary and vehicle theft (Home Office 1988a, 1990a). Also, the downgrading of the offense of unauthorized taking of a motor vehicle to a nonindictable offense (in the Criminal Justice Act 1988) encouraged sentencers to treat it as a relatively trivial offense and to use non-custodial penalties. The Criminal Justice Act 1991, implemented on October 1, 1992, discouraged the use of custody for nonviolent offenses, generally prevented sentencers from taking account of previous convictions or of more than two current offenses, required that persons age 17 should be dealt with as juveniles rather than as adults, and reduced maximum prison sentences for nonresidential burglary and theft. Wilson (1997) argued that up to and including this Act (which greatly reduced the ability of sentencers to pass custodial sentences) Home Office policy makers were primarily concerned with minimizing or (preferably) reducing the prison population. However, the 1991 Act also greatly reduced the use of suspended prison sentences (requiring that they could only be given in exceptional circumstances), and this arguably might have led to an increase in (unsuspended)prison sentences. Home Office policy changed in May 1993 when Michael Howard became Home Secretary. Insisting that "prison works" he encouraged judges and magistrates to make more use of custodial sentences and introduced new laws to facilitate this. For example, the Criminal Justice Act 1993 repealed the provisions in the Criminal Justice Act 1991 that prevented sentencers from taking into account previous convictions or more than two current offenses (with effect from August 16, 1993) and doubled the maximum custodial sentence for persons age 15 to 17 from 1 year to 2 years (from February 3, 1995). As mentioned above the Crime (Sentences) Act 1997, implemented on October 1, 1997, required an automatic life sentence (unless there were exceptional circumstances) for a person age 18 or older convicted for the second time for a serious violent or sexual offense (including homicide, rape, assault causing grievous bodily harm and robbery involving the use of real or imitation firearms). In addition, it required a minimum sentence of 3 years imprisonment (unless there were exceptional circumstances)for a third conviction for residential burglary; however, as all three of these convictions had to occur after December 1, 1999, no offenders were affected by this in 1999. Percentage of sentence served in custody The percentage of a burglary sentence that was served in custody decreased from 58% in 1981 to 44% in 1987 and then stayed constant up to 1997, before increasing to 53% in 1999. Similarly, the percentage of a vehicle theft sentence that was served in custody decreased from 60% in 1981 to 44% in 1987 and then stayed constant up to 1997 before increasing to 51% in 1999. The percentage of an assault sentence that was served in custody decreased from 56% in 1981 to 46% in 1987 and then stayed constant up to 1997 before increasing to 52% in 1999. In contrast, the percentage of a robbery sentence that was served in custody stayed constant from 1981 to 1997 (mean 49%) before increasing to 58% in 1999 Similarly, the percentage of a homicide sentence that was served in custody stayed constant from 1981 to 1997 (mean 45%) before increasing to 58% in 1999. The percentage of a rape sentence that was served in custody stayed constant from 1981 to 1993 (mean 51%) but then increased to 64% in 1999. The percentage of a sentence that was served increased over time for robbery, rape and homicide, and decreased over time for burglary, vehicle theft and assault. The decreases in the 1980's in the percentages of sentences served in custody might be attributed to two changes designed to limit the prison population. In 1981 a prisoner became eligible for parole after serving one- third of the sentence and generally had to be released after serving two-thirds (unless the person behaved badly in prison). The minimum time that had to be served before a prisoner became eligible for parole was 12 months, and prisoners were typically paroled for the last 8 months or so of their sentences. In July 1984, the minimum time that had to be served before parole eligibility was decreased from 12 to 6 months, and a near-automatic parole system was introduced for short-sentence inmates (Home Office 1988b, p.4). In August 1987 the amount of remission for sentences of 12 months or less was increased from one-third to one-half, causing an immediate mass release of about 3,500 prisoners (Home Office 1989, p.19). The Criminal Justice Act 1991 introduced the present system on October 1, 1992. All prisoners serving under 4 years are automatically released after serving half their sentences, while prisoners serving 4 years or more become eligible for release on parole after serving half their sentences and must be released after serving two-thirds (unless they behave badly in prison). Average sentence length The average length of custodial sentences for burglary increased irregularly, from 10.9 months in 1981 to 19.3 months in 1999. The average sentence length for vehicle theft showed no clear trend over time, although it was lowest in 1991. The average sentence length for robbery increased from 27 months in 1981 to 41 months in 1991, and then stayed tolerably constant. The average sentence length for assault increased from 10.8 months in 1981 to 17.8 months in 1991, but then decreased to 14.0 months in 1999. The average sentence length for rape almost doubled, from 40 months in 1981 to 77 months in 1999. The average sentence length for homicide increased from 155 months (12.9 years) in 1981 to 217 months (18.1 years) in 1991 and then stayed tolerably constant until decreasing to 180 months (15 years) in 1999. However, because of life sentences, the estimation of sentence length for homicide cases is problematic (see above). The average sentence length increased markedly over time for all offenses except vehicle theft. Average time served The average time served in custody after sentence for burglary increased irregularly, from 6.3 months in 1981 to 10.2 months in 1999. The average time served for vehicle theft decreased from 4.8 months in 1981 to 2.8 months in 1991, but then increased to 4.2 months in 1999. The average time served for robbery increased irregularly from 13 months in 1981 to 23 months in 1999. The average time served for assault peaked in 1991 at 8.2 months but otherwise stayed tolerably constant. The average time served for rape more than doubled, from 20 months in 1981 to 49 months in 1999. The average time served for homicide increased from 69 months (5.8 years) in 1981 to 105 months (8.8 years) in 1993, but then stayed tolerably constant. The average time served increased markedly over time for all offenses except vehicle theft. The average time served decreased over time for vehicle theft (r = -.57). The average time served for vehicle theft decreased in 1981-1993 (r = -.96) but increased in 1993-1999 (r = .97). There were two main reasons why the average time served and average sentence length for homicide increased in England. First, the number of murder convictions (carrying a mandatory life sentence) increased, whereas the number of manslaughter (including infanticide) convictions stayed constant. For example, in 1981, 126 offenders were convicted for murder and 262 for manslaughter, whereas in 1999 252 offenders were convicted for murder and 264 for manslaughter. Second, the average time served by life-sentence prisoners increased, from 126 months in 1981 to 160 months in 1999. Average time served per conviction The average time served per conviction for burglary increased slightly up to 1993 but then increased considerably, from 91 days in 1993 to 192 days in 1999. The average time served per conviction for vehicle theft decreased from 37 days in 1981 to 12 days in 1991, but then increased back to 37 days in 1999. The average time served per conviction for robbery increased from 293 days in 1981 to 434 days in 1991, and then stayed tolerably constant up to 1997 before increasing to 508 days in 1999. The average number of days served per conviction for assault almost tripled, from 24 in 1981 to 66 in 1999. The average number of days served per conviction for rape increased similarly, from 545 (18 months) in 1981 to 1,438 (47 months) in 1999. The average number of days served per conviction for homicide increased substantially, from 1,770 (58 months) in 1981 to 2,927 (96 months) in 1999. The average days served per conviction increased markedly over time for all offenses except vehicle theft. For vehicle theft the average days served per conviction decreased markedly in 1981-93 (r = -.88) and increased markedly in 1993-99 (r = .96). Average time served per offender The average days served per burglary offender decreased from 1.3 in 1981 to 0.5 in 1993, but then increased to 1.4 in 1999. The average days served per vehicle thief collapsed from 1.9 in 1981 to 0.2 in 1991, but then increased to 0.6 in 1999. The average days served per robber increased from 2.9 in 1981 to 4.5 in 1991, but then decreased back to 2.9 in 1999. The average days served per assaulter fluctuated considerably over time, with the lowest figure in 1995 (0.7) and the highest in 19991`. The average days served per rapist decreased from 148 in 1981 to 94 in 1993, but then showed no clear trend. The average days served per homicide offender increased from 1,117 (37 months) in 1981 to 1,841 (60 months) in 1999. The number of days served per offender decreased markedly over time for vehicle theft (r = -.79) and rape (r = -.69), while it increased markedly for homicide (r = .89). The number of days served per offender for burglary and vehicle theft decreased markedly in 1981-93 and increased markedly in 1993-99. Explaining the results According to the BCS serious property crimes (residential burglary and vehicle theft) increased markedly up to 1993 and then decreased markedly. Very similar trends were seen for police-recorded residential burglary and vehicle theft. According to the BCS assault increased markedly up to 1995 and then decreased, while in police-recorded crimes, assault increased up to 1997 and then decreased. In contrast, robbery increased steadily in the BCS and in police records during the whole period from 1981 to 1999. What hypotheses might be proposed and tested to explain these crime trends? It is extremely difficult to test explanations of changes in national crime rates over time. This is because of the difficulty of controlling statistically for all possible influences on crime rates. A great deal is known about risk factors for offending by individuals, which include individual factors such as impulsivity and low intelligence, family factors such as antisocial parents, poor parental supervision, large family size and disrupted families, socioeconomic factors such as low family income and unemployment, delinquent peers, high delinquency-rate schools, criminal areas and criminal opportunities (for example, Farrington, 1998, 1999). Little is known about differential effects of these factors on different types of crimes, because of the prevailing belief that offending is primarily versatile. Nevertheless, it is possible in studies based on individuals to control for numerous explanatory variables and to determine the time ordering of different explanatory variables and offending. In studies based on countries, many risk factors (for example, impulsivity, parental supervision) are not measured repeatedly at the national level. Even where there are repeated national data on a risk factor (such as unemployment), the method of measurement may change over time, and the national variable, for example the claimant count, may not accurately reflect the theoretical construct of unemployment. There are also great problems of causal lag and causal order caused by the (usually) annual data. After what time lag, for example, is an increase in the unemployment rate likely to cause an increase in the crime rate? If the time lag was short, such as 1 month, then it would not be reasonable to investigate how far the unemployment rate in one year predicted the crime rate in the next year. On the other hand, if the unemployment rate in one year is used to predict the crime rate in the same year, this raises problems of causal order. In general, changes in survey crime rates were highly correlated with changes in recorded crime rates. Survey and recorded crime rates correlated .91 for burglary, .97 for vehicle theft, .94 for robbery and .67 for assault. The correlation was lower for assault because the survey rate showed a greater decrease after 1995; between 1981 and 1995, the correlation for assault was .88 (Langan and Farrington 1998, p.13). Survey crime rates are generally considered to be more accurate. Since most crimes tended to increase over time, changes in most crimes tended to be correlated. The lowest correlations were between survey vehicle theft and survey robbery (r = -.03) and between survey vehicle theft and recorded rape (r = .10), because the survey vehicle theft rate decreased substantially after 1993. In recent years the most important attempts to explain and predict changes in (recorded) crime rates in England have been carried out by economists interested in the relationship between crime and economic prosperity. For the period 1950-87 Field (1990) found that year-to-year changes in burglary, robbery, and theft of vehicles were negatively correlated with year-to-year changes in personal consumption (the average personal expenditure per capita, adjusted for inflation). However, year-to-year changes in violence and sex crime rates were positively correlated with year-to-year changes in personal consumption. Field (1990) concluded that personal consumption was a more important correlate of crime than were the unemployment rate or the Gross Domestic Product (the value of all goods and services produced in the U.K. economy for final consumption) per capita, adjusted for inflation. Personal consumption comprises about half of GDP; the remainder is government expenditure, investment by firms and net exports. At the time this research was carried out, the Conservative government were extremely concerned to deny that there was any relationship between increases in unemployment and increases in crime. In attempting to explain these results, Field (1990) suggested that increases in prosperity meant that people in marginal economic groups were better able to obtain income legitimately and hence had less need to commit property crimes in order to obtain income. However, increases in prosperity also meant that people in marginal economic groups went out more and drank more alcohol, both of which led to increases in violence and sex crimes. Indeed he showed that year-to-year increases in the amount of beer consumed were significantly positively related to year-to-year increases in the rate of violent crime. Surprisingly, Field (1990) found that the year-to-year changes in burglary and robbery were positively related to year-to-year changes in personal consumption 2-3 years before. He therefore suggested that the initial negative effect of changes in prosperity on changes in property crime was followed by a "bounce back" 2-3 years later. Over 5-year periods, the growth in personal consumption was not at all correlated with the growth in property crime. However, changes in personal consumption 2 to 3 years before were still positively associated with changes in violence and sex crimes. The short-term effects of increasing prosperity were similar to the longer-term effects of increasing prosperity for violent crime (unlike property crime). Field (1990) also found that year-to-year increases in the number of young males (age 10 to 29) were positively related to year-to-year increases in all crimes except robbery. Increases in police strength were negatively related to increases in theft of vehicles and sex offenses, and positively related to increases in violence. Increases in the police clearance rate in one year predicted decreases in violence and theft of vehicles in the next year. Increases in car ownership correlated with increases in theft of vehicles, and increased unemployment in one year predicted increased violence in the next year. Pyle and Deadman (1994) attempted to replicate Field's conclusions on property crime (including burglary and robbery) using data that they collected from 1946 to 1991. They concluded that personal consumption and GDP (negatively related to changes in crime) and unemployment (positively related to changes in crime) were essentially interchangeable. They also found that decreases in the conviction rate per recorded offense were correlated with increasing crime rates. Using Pyle and Deadman's (1994) dataset, Hale (1998) aimed to predict both year-to-year changes in crime rates and long-term levels of crime. In agreement with Field (1990), Hale found that year-to-year changes in burglary were negatively related to changes in personal consumption, as well as to police numbers and conviction rates per recorded offense. However, year-to-year changes in robbery were positively related to changes in personal consumption in the previous year, and to changes in the unemployment rate, and negatively related to changes in police numbers. Long-term burglary and robbery levels were positively related to personal consumption; all were increasing over this time period. There have been several other studies of crime and the business cycle in England. For 1988-96 Witt and his colleagues (1999) found that year-to-year changes in burglary and vehicle crime were positively correlated with year-to-year changes in the unemployment rate and in the number of cars per capita, and negatively related to the number of police per capita in the previous year. Dhiri and his colleagues (1999) found that long-term trends in burglary were positively correlated with personal consumption and with the number of young males in the population. Correlations between survey and recorded crime rates and previously identified key indicators (per capita) are given: number of young males in the population, number of males unemployed, personal consumption (consumer expenditure), GDP, beer consumption, number of vehicles, and police strength. Personal consumption and GDP were adjusted for inflation. In addition, correlations between crime rates and the year-to-year change in GDP are shown. All the prior studies were based on recorded crime rates, but survey crime rates are arguably more accurate. We did not attempt to do any complex econometric analyses. The data were obtained from the Office of National Statistics (2000b, 2001). We focus especially on correlations greater than about .5. Crime rates were negatively correlated with the percentage of the population who were young males. The unemployment rate was positively correlated with burglary and vehicle theft rates but weakly correlated with the other four types of crimes. Inflation-adjusted personal consumption (consumer expenditure) per capita was positively correlated with all types of crimes except survey vehicle theft and recorded burglary and vehicle theft. The inflation-adjusted GDP per capita was positively correlated with all types of crimes except survey vehicle theft. In general, personal consumption and the GDP per capita tended to increase over time, whereas survey vehicle theft increased a great deal and then decreased a great deal. The year-to-year change in GDP was generally weakly related to crime rates; it was most strongly related to survey vehicle theft (r = .50). Beer consumption per capita was negatively related to all types of crimes. It was least strongly correlated with survey vehicle theft (r = -.27). The number of vehicles per capita was least strongly correlated with vehicle theft (r = .21) but positively correlated with all other types of crimes. In general, the number of vehicles per capita increased steadily over time, and so it was positively correlated with types of crimes that increased steadily over time. Finally, police strength per capita was most strongly correlated (positively) with burglary and vehicle theft, and not strongly related to the other four types of crimes. In most cases these results are not as expected. Contrary to expectation, survey crime rates did not increase as the number of young males increased, did not increase as beer consumption increased, and did not decrease as police strength increased. There was no evidence that increasing prosperity caused any decrease in crime, and vehicle theft did not increase with the number of vehicles available to be stolen. The most expected result is that burglary and vehicle theft increased as the unemployment rate increased. Previous publications (Farrington and Langan, 1992; Farrington and Wikström, 1993; Farrington, Langan, and Wikström 1994; Langan and Farrington, 1998) investigated correlations between changes in crime rates and changes in the probability and severity of legal punishment. Of course, it must be realized that these correlations are based on only a few years (8 for survey crimes and 19 for recorded crimes in this chapter)and that correlation does not necessarily indicate causation. This report shows correlations between the survey crime rate and different measures of the probability and severity of legal punishment; similar correlations are presented for recorded crime rates. We believe that our measures of the probability of punishment (for example, the number of convictions per 1,000 offenders) are more accurate than those used previously for example, the ratio of the number of convictions to the number of recorded crimes). The percentage of crimes reported and the probability of recording a reported crime were not consistently related to survey crime rates. Nor was the conviction rate per capita, which was positively related to robbery, rape, and homicide but negatively related to burglary, vehicle theft, and assault. However, the number of convictions per 1,000 offenders was highly negatively related to all types of survey and recorded crimes. The lowest correlation was with homicide (r = -.43). The custody rate per capita and the probability of custody following a conviction were not consistently related to survey or recorded crime rates. However, the custody rate per 1,000 offenders was highly negatively correlated with all four survey crimes and with four of the six recorded crimes. The average sentence length average time served and percentage of sentence served in custody were not consistently related to survey or recorded crime rates. Nor was the average number of days served per conviction. However, the average number of days served per offender was highly negatively correlated with all four survey crimes and with three of the six recorded crimes. It seems clear that the number of convictions per 1,000 offenders was most consistently (negatively) related to survey and recorded crime rates. The results obtained with the custody rate per 1,000 offenders and the average number of days served per offender are probably consequences of the fact that both measures are a function of the number of convictions per 1,000 offenders. Correlations between survey crime rates and the number of convictions per 1,000 offenders might be considered problematic because the number of convictions per 1,000 offenders is not independent of the number of survey crimes. Nevertheless, it is not necessarily true that these two measures must be negatively correlated. If the number of survey crimes increased by 10%, for example, the number of convictions per 1,000 offenders could decrease by 10% (if the number of convictions stayed constant), stay constant (if the number of convictions increased by 10%) or increase by 10% (if the number of convictions increased by 20%). The recorded crime rate is logically independent of the number of convictions per 1,000 offenders, although of course survey and recorded crime rates are correlated. If the probability of being convicted influences the behavior of potential offenders, the number of convictions per 1,000 offenders should predict the crime rate rather than the reverse. "Conviction Rate/Offense A" indicates predictive correlations with the crime rate in one year predicting the number of convictions per 1,000 offenders in the next year ("crime first"). "Conviction Rate/Offense B" indicates predictive correlations with the number of convictions per 1,000 offenders in one year predicting the crime rate in the next year ("crime second"). For all four correlations with survey crime rates, the negative correlations between convictions and crimes were greater for crimes predicting convictions rather than for the reverse. For recorded crime rates, this was true for burglary, vehicle theft and robbery but not for assault. The "crime first" and "crime second" correlations were very similar for assault and negligible for homicide. These results suggest either that changes in the probability of punishment have a quick effect (in less than one year) on the crime rate or that they do not have a deterrent effect. Correlations exist between survey and recorded crime rates and the custody rate per 1,000 offenders and the number of days served per offender for the "crime first" (A) and "crime second" (B) conditions. In 20 tests negative correlations were higher in the "crime first" (A) condition in 7 cases, lower in 5 cases, and similar (differing by .04 or less) in 4. In the other four cases neither correlation was highly negative as predicted. Therefore, there was no consistent tendency for the probability of punishment in one year to predict the crime rate in the next year more than the reverse. It is very unlikely that any single factor will explain changes in crime rates over time. Different factors may influence different types of crimes in different time periods, possibly depending on the magnitude of change in the explanatory factor. Large changes in short time periods might be expected to have greater effects. For example, the dramatic decrease in the probability of conviction for vehicle theft (from 52 convictions per 1,000 offenders in 1981 to only 12 in 1993) might be expected to have a relatively large effect on vehicle theft. However, negative correlations between the probability of conviction and the crime rate were less for vehicle theft than for burglary, robbery, or assault. In conclusion, this chapter presents basic data on crime rates and punishment rates that needs to be explained by criminological theories. The correlations reported in this section are thought-provoking but do not provide a convincing explanation of the crime trends. They do not prove convincingly that either the probability or the severity of legal punishment have a deterrent effect. Nor do they prove any causal effects of prosperity or the age structure of the population on crime rates. In order to obtain better data, offenders should be tracked longitudinally through the criminal justice system. Better data is especially needed on the crucial measure of the average number of offenders per offense, which is needed to link up offenses and offenders. In future, it would be desirable to expand these analyses to more stages of the criminal justice system (for example, arrest and prosecution). More research is needed especially to explain why residential burglary and vehicle theft increased so much in 1981-1993, then decreased so much in 1993-1999. References Barclay, Gordon C. (1993) Personal communication. Barclay, Gordon C. (2001) Personal communication. Budd, Tracey. (2001) Personal communication. Chapman, Becca, and Stephen Niven. (2000) A Guide to the Criminal Justice System in England and Wales. London: Home Office. Dhiri, Sanjay, Sam Brand, Richard Harries and Richard Price. (1999) Modelling and Predicting Property Crime Trends in England and Wales. London: Home Office. Farrington, David P. (1989) "Self-Reported and Official Offending from Adolescence to Adulthood." In Cross-National Research in Self- Reported Crime and Delinquency, Malcolm W. Klein, ed. Dordrecht, Netherlands: Kluwer. Farrington, David P. (1992) "Trends in English Juvenile Delinquency and Their Explanation." International Journal of Comparative and Applied Criminal Justice 16: 151-163. Farrington, David P. (1998) "Predictors, Causes and Correlates of Male Youth Violence." In Youth Violence, Michael Tonry and Mark H. Moore, eds. Chicago: University of Chicago Press. Farrington, David P. (1999) "Conduct Disorder and Delinquency." In Risks and Outcomes in Developmental Psychopathology, edited by Hans-Christoph Steinhausen and Frank C. Verhulst. Oxford: Oxford University Press. Farrington, David P. and Patrick A. Langan. (1992) "Changes in Crime and Punishment in England and America in the 1980s." Justice Quarterly 9: 5-46. Farrington, David P., Patrick A. Langan, and Per-Olof H. Wikström. (1994) "Changes in Crime and Punishment in America, England and Sweden between the 1980s and the 1990s." Studies on Crime and Crime Prevention 3: 104-131. Farrington, David P. and Per-Olof H. Wikström. (1993) "Changes in Crime and Punishment in England and Sweden in the 1980s." Studies on Crime and Crime Prevention 2: 142-170. Field, Simon. (1990) Trends in Crime and Their Interpretation. London: Her Majesty's Stationery Office. Grace, Sharon, Charles Lloyd and Lorna J. F. Smith. (1992) Rape: From Recording to Conviction. London: Home Office. Hale, Chris. (1998) "Crime and the Business Cycle in Post-War Britain Revisited." British Journal of Criminology, 38: 681-698. Home Office. (1985) The Cautioning of Offenders (Circular No.14/1985). London: Home Office. Home Office. (1987) Prison Statistics, England and Wales, 1986. London: Her Majesty's Stationery Office. Home Office. (1988a) Punishment, Custody and the Community. London: Her Majesty's Stationery Office. Home Office. (1988b) Report of the Parole Board, 1987. London: Her Majesty's Stationery Office. Home Office. (1989) Prison Statistics, England and Wales, 1988. London: Her Majesty's Stationery Office. Home Office. (1990a) Crime, Justice and Protecting the Public. London: Her Majesty's Stationery Office. Home Office. (1990b) The Cautioning of Offenders (Circular No. 59/1990). London: Home Office. Home Office. (1993) Criminal Statistics, England and Wales, 1991. London: Her Majesty's Stationery Office. Home Office. (1996) Criminal Statistics, England and Wales, Supplementary Tables, 1995. London: Government Statistical Service. Home Office. (1998a) Criminal Statistics, England and Wales, 1997. London: The Stationery Office. Home Office. (1998b) Criminal Statistics, England and Wales, Supplementary Tables, 1997. London: Government Statistical Service. Home Office. (2000a) Criminal Statistics, England and Wales, 1999. London: The Stationery Office. Home Office. (2000b) Criminal Statistics, England and Wales, Supplementary Tables, 1999. London: Government Statistical Service. Hough, Mike, and Pat Mayhew. (1983) The British Crime Survey: First Report. London: Her Majesty's Stationery Office. Irving, Barrie, l. and Ian K. MacKenzie. (1989) Police Interrogation. London: Police Foundation. Kershaw, Chris, Tracey Budd, Graham Kinshott, Joanna Mattinson, Pat Mayhew and Andy Myhill. (2000) The 2000 British Crime Survey, England and Wales. London: Home Office Statistical Bulletin (18/00). Langan, Patrick A. and David P. Farrington. (1998) Crime and Justice In the United States and in England and Wales, 1981-96. Washington, D.C.: Bureau of Justice Statistics. Lavin, Duncan. (2001) Personal communication. Mattinson, Joanna. (1999) Personal communication. Mattinson, Joanna. (2000) Personal communication. Mayhew, Pat. (1996) Personal communication. Mayhew, Pat. (1997) Personal communication. Mirrlees-Black, Catriona. (1996) Personal communication. Mirrlees-Black, Catriona, Tracey Budd, Sarah Partridge and Pat Mayhew. (1998) The 1998 British Crime Survey. London: Home Office Statistical Bulletin (21/98). Office for National Statistics. (2000a) Britain 2001: The Official Yearbook of the United Kingdom. London: The Stationery Office. Office for National Statistics. (2000b) Estimated Resident Population at Mid-1999 by Single Year of Age and Sex, England and Wales. Titchfield, Hampshire: Office for National Statistics. Office for National Statistics. (2001) Annual Abstract of Statistics, United Kingdom. London: the Stationery Office. Povey, David, and Julian Prime. (1999) Recorded Crime Statistics, England and Wales, April 1998 to March 1999. London: Home Office Statistical Bulletin (18/99). Pyle, David J., and Derek F. Deadman. (1994) "Crime and the Business Cycle in Post-War Britain." British Journal of Criminology 34: 339-357. Stevens, Roger. (1999) Personal communication. Stevens, Roger. (2000) Personal communication. Wilson, James Q. (1997) "Criminal Justice in England and America." The Public Interest 126:3-14. Witt, Robert, Alan Clarke and Nigel Fielding. (1999) "Crime and Economic Activity: A Panel Data Approach." British Journal of Criminology 39: 391-400. Acknowledgment The authors are very grateful to Maureen Brown for efficient word processing. They are also grateful to Margaret Ayres, Gordon Barclay, Tracey Budd, Billy Burns, Patrick Collier, LizAnne Dowds, Graham Kinshott, Duncan Lavin, Joanna Mattinson, Pat Mayhew, Catriona Mirrlees-Black, Chris Nuttall, Andrew Percy, Roger Stevens, David Thomas, and others from the Home Office for providing helpful information over the years. Authors David P. Farrington, O.B.E. (Officer of the British Empire) (see page xiv of the Introduction for biographical profile). Darrick Jolliffe (see page xiv of the Introduction for biographical profile). United States Goal This chapter describes trends in crime in the United States between 1981 and 1999, and trends in justice system performance between 1981 and 1996, and explores possible explanations for them. Six offenses are investigated: residential burglary, motor vehicle theft, robbery, serious assault, rape, and homicide. The term "trends in crime" means trends based on two distinctly different ways of measuring the volume of crime in the United States: trends in "survey" crime rates for four offenses (robbery, assault, residential burglary, motor vehicle theft), so called because they are based on victim surveys trends in police-recorded crime rates for six offenses (homicide, rape, robbery, assault, residential burglary, motor vehicle theft) "Trends in justice system performance" is defined by 13 different measures that capture the movement of persons into and out of the justice system. Five of the 13 can be thought of as measures of "risk of punishment," and 4 can be thought of as measures of "punishment severity": Risk of punishment measures trends in arrest rate per 1,000 offenders trends in conviction rate per 1,000 arrested offenders trends in conviction rate per 1,000 offenders trends in custody rate per 1,000 offenders trends in days served per offender Severity of punishment measures trends in percent custody per conviction trends in average sentence length trends in average time served trends in percent of time served Other measures trends in percent of crimes reported to police trends in percent recorded of reported trends in conviction rate per 1,000 population trends in custody rate per 1,000 population These 13 justice system performance measures were selected for a combination of reasons: national statistics are available to compute them; relatively speaking, they are frequently available in other countries, making international comparisons possible; together they provide a fairly comprehensive overview of the functioning of the justice system. The rationale for selecting particular years for study was similar: the selected years are ones for which the needed national data exist; and other countries have comparable data for these years or for years close to them. United States Description The United States is an ethnically diverse country, with a 1999 resident population of 272.7 million. Between 1981 and 1999 the U.S. population grew by 19%, in part due to large increases in immigration. Over that period non-whites went from 14% to 18% of the total population, and Hispanics' share went from 6% to 11%. By 2010 Hispanics are expected to outnumber blacks and become the Nation's largest minority. By 2050 Hispanics are expected to comprise a fourth of the U.S. population. Life expectancy continues to rise in the United States. A person born in 1981 had a life expectancy of 74 years; one born in 1999, 76 years. The population of the United States is aging. Median age rose from 30 years in 1981 to 36 in 1999. Persons in the high crime-prone age category -- ages 13 to 24 -- are a declining percentage of the population: 27% in 1981 versus 20% in 1999. Marital status of the population has changed since 1981. Divorced persons made up 6% of the population ages 18 years and over in 1981, and 10% in 1999. Never-married persons went from 20% to 24%. Births to unmarried women made up a growing share of all births in the United States from 1981 to 1999, going from 18% to about 32% of the total. Also during this period the abortion rate generally fell. In 1981 the rate per 1,000 women 15 to 44 years of age was 29, compared to about 23 in 1999. Educational attainment in the United States continues to rise. In 1981 16% of persons 25 years old and over had completed 4 years of college or more, compared to about 25% in 1999. Over the study period the unemployment rate generally fell, from 7.6% in 1981,to 4.2% in 1999, while no particular trend was evident in the poverty rate: 11% below the poverty level in 1981 and about that same percentage in 1999. The criminal justice system The United States is made up of 50 States and thousands of cities and counties within those States. The vast majority of these cities and counties have their own criminal justice system, and each of the 50 States has its own justice system. In addition there is a Federal government that also has its own justice system. What this all means is that, by and large, each city and county, each State, and the Federal government has its own police, its own prosecutors, and its own courts, and, less commonly, its own correctional agencies and its own laws. No one level of government has overall responsibility for justice administration. The Federal government does try to encourage conformity to certain standards among the many thousands of police agencies, prosecutors, courts, and correctional agencies that together make up the criminal justice system in the United States. But the Federal government's authority is severely limited. For example, it cannot stop a State from, say, releasing prisoners before they have served at least 85% of their sentence. But the Federal government can and does offer financial incentives to States for keeping prisoners confined until they have served 85%. If all criminal justice in the United States were administered by a national criminal justice system, governed by a single set of national laws, one of the chapter's aims -- to investigate changes over the study period in justice system performance by possible reference to changes in the law -- would be manageable. But when there are least 51 governing sets of laws (the la