A FIRST ANALYSIS OF THE UNITED NATIONS DATA SET ON CRIME TRENDS AND THE OPERATIONS OF CRIMINAL JUSTICE SYSTEMS

I: INTRODUCTION AND BACKGROUND.

The United Nations has collected data on crime and criminal justice since the mid 1970s, but little systematic use has been made of the data, with some exceptions. The data were collected in a series of five sweeps, called by the UN "surveys". Thus the first survey covered 1970-74, the second 1975-80, the third 1980-86, the fourth 1986-90 and the fifth 1990-94. In this report the word "sweep" is used, so that the term "survey" can be retained for the activity as a whole. The sixth sweep, covering 1995-97 is being administered in early 1999.

The data set has been reconfigured from the form in which it was earlier on the Internet into a standard format wherein each file is one year, and all variables have unique identifying numbers, as do all cases (countries). The objective of this exercise is to make the data set available to all interested scholars in a format which makes it easy for them to carry out their own analysis on whatever variables and cases interest them the most.

II: THE QUALITY OF THE DATA

One major concern in the analysis of any data set must be with the quality of the data, traditionally evaluated by validity, the accuracy with which the data relate to what is being measured in the external world, and reliability. There are two kinds of reliability, external, that is the degree to which the data in a set under consideration correspond to other data sets which purport to measure the same parameter, and internal, that is the accuracy of the apparent relationship of one datum to the other data within the same set. This analysis is concerned with internal reliability only. In order to facilitate the monitoring of this question, in each sweep the first year requested was the same as the last year of the preceding sweep

An exploratory analysis of the years for which data were requested as the last year of a given sweep and, five years later, the first of the next sweep, hereafter referred to as the "overlapping" years, has been carried out. Each of the cells or fields which provides the figure for a particular variable and case for the same year should show identical numbers, whenever it was collected. As Table 1 (below) shows, in the variables chosen to test this situation, there were very few perfect "matches", and even if figures which are less than 10% different for the same field are allowed to count as a match, still more than half the figures in some categories of variables do not form matches. There are marginally more matches between the fourth and fifth sweeps, but that is not consistently the case. Recorded crimes by crime type was chosen because it is a figure perhaps most often used in international comparisons. One form of analysis which would give a comprehensive insight into the reliability of the data would be to repeat this procedure for all variables, so that a classification of variables could be made into those which have many matches and those which have relatively few; each of these categories could be further classified into a rank order. That would, however, be a major task in itself. The important point is that the data for cases and variables being used for any analysis are subjected to such tests to establish a confidence level before that analysis is undertaken.

Table 1 - percentage of entries in overlapping years with matching data (within 10%)

Numbers such as these suggest that the external validity of the data may be suspect. In particular the fact that for two of the major categories of crime, assault and theft, not even matches within 10% were recorded in one third of all cases is decidedly disconcerting. There is an argument to be made that such apparent inconsistencies between sweeps may not indicate lack of validity or reliability within sweeps. Each set of data was presumably collected by different people at five yearly intervals, and he/she may have used different categories of files from those used by their predecessor. However, the comparison of rates of change over the years, discussed in the following paragraphs, lend only limited support to that view. The overlapping years discrepancies suggest that the internal reliability of the data may be suspect, and raise even more serious questions over the external validity.

Someone wishing to use some of the data set for analysis, therefore, would be well advised to carry out the procedure, described in Technical Annex 1, to establish whether the particular variables and cases of interest demonstrate a high or low level of discontinuity in respect of the overlapping years. While it would be possible to analyse every variable for every case for which data exist and present the results in some comprehensive report thereof, it would be a very long and tedious report, and, more importantly, the great bulk of it would be irrelevant to any given reader. It seems both more efficient and more constructive, therefore, to present and discuss some examples in the following paragraphs, and to facilitate the carrying out of the analysis by different individuals on the variables and cases of interest to them by means of the procedure described.

The validity of the data cannot be checked further without going back to the countries and asking them both to reopen old files and then cross check the numbers contained therein. Such a procedure is almost certainly not feasible, and would cost much more in time and effort than the information so generated would be worth.

The internal reliability of the data can, however, be checked further. The main method used has been the comparison of the changes in values for each variable with earlier figures for that variable. This can be done in two ways. In the first, the figure for one variable is compared with that for the same variable the year before, and the extent of difference recorded as a percentage change, referred to hereafter as "adjacent years". In the second, the figure for a particular year is compared with the figure for a base year, which is held constant for all subsequent years, hereafter called "lagged" years. Both sets of the numbers so derived represent the degree of change moving across time, but "time" is conceptualized differently. In the first instance, it is a creeping relative variable, and in the second an absolute value against a constant base. Each set of results shows the change in the rate of change of the variable in question. Adjacent years show the changes more clearly and sharply; lagged years show the overall trends across many years. It is therefore worthwhile generating both sets of numbers, as described and illustrated in Technical Annex II

As the first sweep had so many missing variables and cases, the exercise was started with 1975 as the first year, and as the base year for the lagged method. Some results are shown in Charts 1-6. Charts 1-4 show the data for Singapore and Sweden for the whole period of the second through fifth sweeps, for the number of robberies recorded by the police and the number of convictions for robbery recorded by the courts. These data have been analyzed in both the ways described above, adjacent years and lagged years. In both charts it is important to bear in mind at all times that these lines represent changes in the rate of incidence of the offence or conviction, and NOT the incidence itself in any one year.

The mean for all countries submitting complete data in respect of this variable was also computed, and is shown by the thick line, for adjacent years. As the lagged years line is cumulative, the mean quickly climbs above the upper limit of the scale which is appropriate to demonstrate the variations in these two countries. The lagged charts therefore do not show a mean. Both the countries shown are at or below the mean, especially for recorded offences. That suggests that these two countries are not outliers, and that observations made in respect of the variance to be found in their case are probably conservative for countries over all. Given the relative reliability of robbery as a crime type, as described in the next section, this suggests that the data in general cannot be assumed to be internally consistent, and that therefore their reliability is questionable, with the validity in consequence equally questionable.

One important difference between the adjacent and lagged years charts is that in the former the graph line can turn downwards, if the rate of change is lower than the previous year. For the lagged years, that can occur only when the rate of change is for a given year is less than that for 1975-76. A flat line for adjacent years would mean that the rate of change is consistent and constant. A flat line for the lagged years would mean that there has been no change at all in the incidence rate.

The pictures they give of the changes in the recorded incidence of the variable in question differ in that the percentage change per year by adjacent years is less smooth than that by lagged years for recorded robberies; but that difference is not found for the recorded convictions. The changes in the convictions rate seem to be both greater and more frequent than .for the offence rate, which raises the question of why this should be so. The opposite might have been expected to be the case, if there was some form of unrealised stabilising factor in the operations of the prosecution and courts which led to the same proportion of offenders being convicted each year, irrespective of a vacillating input; but these charts suggest that the situation might be the other way round. A possible explanation might be the statistical phenomenon for variation to increase with a decrease in the base number, that is, small numbers show more variation than large numbers, other things being equal; and the numbers of convictions is intrinsically only a quite small proportion of the number of cases recorded. But this is a phenomenon which may repay wider study over other countries and crime types.

Charts 5 & 6 show the results from the same analysis for Chile and England & Wales, in respect of the convictions for robbery. The smoother curve of the lagged years graph is again apparent, as is the fact that the longer established, perhaps more stable industrialized country shows a lower rate of change than the newer country, one which went through a very difficult political time during the period under review. In these examples at least, the variation in the data for the fifth sweep seems even greater than for the earlier sweeps.

The line for the mean is included for two reasons. The first, as mentioned, is to show that the countries selected are not in any way outliers, but rather show considerably less variation than the mean; that is to say, the data from many other countries are considerably less stable. The second is to demonstrate in another way the problem discussed above, of the overlapping years. The years for which the mean goes off the charts (literally) are mostly those of the changeover from one sweep to the next, after which the rate of change seems to settle down to more or less the level of the previous sweep. Within each sweep, the change in (instability of) the data is disconcertingly high. Across sweeps it becomes very much higher.

One of the most important questions which is raised by this examination of the reliability of the data is whether, or to what extent, the apparently questionable reliability of the data is or is not an indicator of the validity or lack thereof. That is, do the case flows in the system actually vary to the extent indicated, or is there a source of imprecision in the recording process. If the data do change from year to year as much as these charts suggest, and if those changes do accurately record changed levels of incidence of a given variable in an agency of the criminal justice system, one of the most widely held beliefs about the nature and operations of criminal justice is brought into question. It is something of a criminological commonplace that the operations of the criminal justice system are stable, that one year is a good predictor of the next in terms of workload, allowing for a general tendency to increase, or, in recent years in some industrialized countries, to decrease a little. It would require a separate and extensive analysis of all the variables for all years to explore that hypothesis thoroughly, but the samples taken do suggest that either the numbers for the system agencies are not as stable as is usually assumed or that the recording of the data introduces apparent fluctuations which are not actually "there", i.e. the data are not as valid as might be wished.

These tables and charts suggest that confidence in the quality of all the data should be tentative and qualified, and even more that the appropriate or justified confidence level is specific to the subset of data being analysed. It is therefore essential that some form of quality evaluation of any given subset be carried out on that subset, so that a confidence level can be assessed. From these few samples, it seems that the question of reliability, and therefore perhaps of the underlying validity, might well vary not only from case to case but from variable to variable. The inference to be drawn from this that while the results of analyses of the data set which have already been carried out may be true, they may also not be true in the sense that it is impossible to be sure they are true, where "being sure" is defined as being above a certain level of confidence. The analysis in this report does not in any way assert that such analyses are incorrect; but rather that, because of the suspect quality of the data, it is not possible to assert that they are correct with what is usually taken in social science as an accepted level of confidence. There is no alternative to any user assessing the reliability of each variable which is used in an analysis, either with the procedures suggested above or alternative measures of evaluation.

The analysis described above was one of the factors which led the author to the view expressed several times in this report that the data can be used for suggesting and refining questions for further study, but not, with any authority, for activities demanding statistical rigour such as hypothesis testing. That is, one of the main appropriate uses of this data set is to identify a next generation of more specific and clearly focussed questions for precise study.

III: ON ANALYSING THE DATA

There are some well known and frequently rehearsed difficulties in respect of the analysis of data on crime and justice across countries, cultures or jurisdictions. The first of these is that of definitions. Different legal codes define crimes in different ways, so that the set of acts which constitute a given crime type in one country may not be identical to the set of acts to which the same label is applied in another. One well known example is "Rape"; another less well known example, but perhaps more illustrative of the point, is that the concept and crime type of "Attempted Homicide", which is used in many countries, is not known in the USA, where it is classified as a "Major Assault". The consequence of this is that any comparison of total homicide figures which include "attempts" with the homicide figures for the USA will have an inbuilt distortion, and the same will occur in reverse in the comparison of assault figures.

The second is that of recording practices. Different police forces, in particular, have different rules for when an event should be recorded as a crime and when not. For instance, in some countries the police are said to be punctilious about recording every theft of a bicycle, whereas the police in other countries with a higher workload of serious crimes and which are less well resourced and perhaps organized might not always record the event, particularly if the bicycle were recovered soon afterwards. The degree and manner in which the agencies do differ across countries is an interesting topic for research, but far beyond the scope of this work. It is simply assumed that such variation does exist, and is likely to affect the numbers for the earlier stages of the criminal justice process considerably. Recording convictions in court and receptions into prison should be more likely to be comparable, because there is less room for variation; but even in that respect some caution may be appropriate.

The third is that of operating practices. In some countries the prosecution stage and process is the locus of the main decisions affecting a case, so that many cases, especially trivial ones, do not appear in the records of those countries until the prosecution stage. Common law and codified, civil law, countries vary also in this respect, and comparisons between the raw numbers of different systems can be risky unless the person making the comparison is familiar with the details of the modus operandi of the system and its implications for the statistical recording thereof. In short, the old scientific adage that understanding a set of data is a necessary precondition for the proper analysis of it applies in this context also.

Fourthly there is a large factual inequality between countries as to their size of population, the make up of the population (for instance, % urban and % rural, % over 60 and % under 25 years of age) and the size of the crime problem, even if prorated by population. In any exercise in even a prorated direct comparison there will be hidden factors affecting the outcome.

Finally, there are a set of problems specifically associated with the category of recorded crime. There is an extensive literature on this topic, and the only observation needed here is that, for the purposes of analysis in this report, the numbers provided by governments are regarded as indicators of the input into, and therefore workload of, the criminal justice system. They are not regarded as accurate statements as to the actual prevalence and incidence of a given crime type in a given jurisdiction, although they may be that. Further information would be needed to validate the figures. It is general criminological wisdom that the less serious, or less obvious, the crime type, the more questionable the officially recorded figures. It is also widely accepted that victim surveys provide more valid data in regard to incidence of most types of crime.

All of these arguments are good reason for caution in regard to a direct comparison of the numbers across countries and jurisdictions.

There is, however, one type of analysis which is not vulnerable to such uncertainties: the use of ratios within countries. While the definition of a crime, or the unit or rules of counting may vary across countries, they do so to any extent within a given country or jurisdiction only relatively rarely, although it does occur. While the assumption that the definition of a certain type of crime is significantly different between two countries is easily supported, the assumption that the definition of the crime type does not change significantly across a number of years within one of the countries is also usually valid, although anyone making use of small differences might be wise to check that there have been no major legislative changes in respect of the crime type in the country(ies) and in the period under study.

Common sense suggests that the categories of data which can most profitably be analyzed are those which are reported most comprehensively. The crime type which is probably the best recorded, in the sense of completed number of cells in the questionnaires returned by the governments, is Rape. It has not been used in the example analyses in the following sections, however, because it is the type of crime which, while not subject to frequent changes in legislation, is most vulnerable to changes in reporting rate by victims and recording practices by the police, as public concern over the offence changes. That is to say, there is probably significant internal inconsistency in the figures for Rape within a country so that even the ratio approach is open to question. As the purpose of the analysis in this report is to demonstrate the nature and feasibility of some types of analysis, one of the most serious types of crime has not been used. The reasons are methodological, and in other contexts the analysis of the figures for Rape would be an appropriate activity. However the research worker would need to check the consistency question with the countries in question if precise significance is to be given to specific changes in the numbers reported.

Rape is both a serious crime and a well reported crime by governments, in the sense that the data cell is rarely left blank, but, as described above, both the validity and the reliability of the figures at the national level are so questionable as to remove any strong confidence in their use for analytic purposes. Homicide as a crime type also has too many variations and sources of uncertainty. The crime type which has been chosen as the most reliable and valid for purposes of example in this report is "Robbery". One indicator that this might be appropriate is that the percentage of overlap in Table 1 is highest for robbery over all. It is serious enough to be reported in most cases to the police, recorded by them and considerable resources are invested in tracing offenders. It is often thought of as a crime against the person, but in fact the ultimate objective is property, and in some national criminal codes it is classified as a crime against property. That is, it bestrides the division between crimes against the person and crimes against property. While the `recorded robberies' must be regarded with the standard caution for that category of variable, it seems from victim surveys to be as accurately recorded as any other, and does not have the sub-divisions which are potential sources of statistical difficulty in the case of homicide, assault and theft.

To return to the ratio approach. The data set is primarily made up of data on the operations of the agencies of the criminal justice system. Each of these can be seen, sequentially, as receiving the input from the preceding agency and creating, as its output, the input for the next stage. The units of count in all these operations are people, and, within one national system, it is feasible to express the values for any stage as a ratio of the preceding or succeeding stage. The most simple form of ratios is a percentage, which is a ratio against base 100. The stage which is selected as the base, and set as 100, can be whichever is the most useful for the particular analysis being undertaken. If a later stage in the criminal process, such as conviction by the courts or reception in to prison, is selected, the figures for the earlier stages will be higher than 100%, unless there is something very unusual about that system.

The figure used for the base in most of the analyses in the next section is that for Recorded Crimes. One very important observation in that regard is that the unit of count is no longer people but events, namely the decision of the official, usually a police officer, to whose attention the offence is first brought to record it formally. In fact, all the units of count are, in strict system theoretic terms, the consequences of decisions made by officials; but that technical nicety does not affect the general nature of the analysis. Thus in the charts of the attrition through the system which make up much of the next section, the first stage, from recorded crime to number of individual people suspected/apprehended for that type of offence, often emerges as a very sharply declining line. The change in the unit of count, from acts/events to people, is one influencing factor, because, even with a 100% clear up rate, repeat offenders would cause some decrease in the `apprehended' stage. The reason for that is that many offences are not cleared up, and quite a lot of offences, especially property crimes, are committed by the same offender. Thus although that slope can be interpreted as a crude measure of clear up rate, its use is most properly within the same jurisdiction to compare either across crime type or across years. If it is used for cross national comparisons, the analysis can be made properly only if the recording practices of each of the countries involved in the comparison are known. Otherwise, especially in respect of relatively common and minor property crimes, such as bicycle theft, those jurisdictions which are punctilious about recording all crimes emerge on the charts as having very low clear up rates, and what is in fact a tribute to their thoroughness can easily be misinterpreted as an indicator of inefficiency. This is, therefore, another instance of how it is necessary to understand the data before drawing specific conclusions about specific countries or agencies.

The ratios of the progress of cases through the agencies of the criminal justice system is most easily plotted and understood as a line graph, and there are several examples in the next section. Its logic is derived directly from, and as a simplified version of, the attrition funnel which was first brought to public attention in the Report of the President's Commission, The Challenge of Crime in a Free Society, in 1967. The attrition funnel can also be plotted by gender and by age, which for this data set means `adult' and `juvenile'. It can therefore show whether males are filtered out of the process at a faster or slower rate than females, and whether juveniles are filtered out faster or slower than adults. These questions, however, can be asked only in respect of "total crimes". The UN questionnaire for the second sweep asked for breakdown by age and crime type, but very few countries were able to answer, and the questions by age, by gender and by crime type were dropped from subsequent sweeps.

In the charts, the values for each stage are constrained by the other values as a matter of practicality, but they are not related logically to the others, in the sense that there is no total to which they have to add up. Ratios can also be plotted as pie charts, and these are the most appropriate display when the purpose is to show how a given amount of something is divided into or between different sub-components. Thus the division of resources between the different agencies of the criminal justice system can be expressed as ratios in a pie chart. These pie charts can be generated using the number of people employed or the budget figures provided by governments. The number of people can include civilian support staff, or just the qualified officials (sworn police officers, district prosecutors, judges etc). Examples are discussed in the next section.

With these two sets of ratios, a simple set of "criminal justice profiles" can be generated for each jurisdiction. The line graphs can show the outline of how the system processes offenders in terms of the ratio carried on to the next stage, by crime type or any combination of crime types; while the pie chart can show the allocation of resources to each agency. Each of these types of display can be repeated for any year, or over longer periods, so that some representation of the basic dynamics of the criminal justice emerges.

IV: SOME INTERPRETATIONS OF THE DATA.

After the rather abstract discussion of the previous section, it is now appropriate to look at some of the results obtained from the analyses along the lines suggested in Section III. First, however, this is an appropriate place to reinforce and explain in detail why this section deals only with examples, and makes no attempt at a comprehensive analysis. The critical factor is the number of possible charts which could be drawn for each country.

There are nineteen categories of crime as shown in Table 1 in Section 11, to which the category of "Total Recorded Crimes" should be added. The data set covers twenty years. If a country submitted a complete return to each sweep, the number of attrition funnel charts which could be created is three hundred and eighty. One chart per year each for differential attrition by gender and age, and two potential pie charts per year on resource allocation would produce a total of eighty more. In reality, no country has submitted complete data, and many of those charts are not feasible. Which charts are feasible can be established only by examining the data set.

Even if only forty countries had returned enough data to allow a half complete set of charts, the resulting report would be very long, very repetitive and rather boring, because in all probability most of the contents would be of no particular interest to any one reader. Very few readers in the USA are likely to be interested in a comparison of the profiles of, say, Denmark and Japan; but Danish and Japanese criminologists might be very interested. The number of charts could then be multiplied by a very large number if comparisons were made of two or more countries, in respect of one or more decision stages of the criminal process. Even if several countries were plotted on the same chart, the number of possible combinations is very large.

The objective of this and the following sections, therefore, is to illustrate, and by the illustrations stimulate readers to carry out the analysis which interests them, by country, crime type or any other variable. The methodology for doing that is given in the Technical Annexes 3, 4 and 5.

The first set of attrition funnels shows the processing of offenders through the system, in the sense of percentages at each stage of the base figure of crimes reported. These charts have been used because they show some possibly interesting similarities and differences in processing patterns between two neighbouring and similar countries. The differences are across crime types and over time.

Charts 7 & 8 show the attrition funnel for Norway in 1982 and 1990 in respect of robberies and theft, and Charts 9 & 10 the same charts for Sweden, for 1982 and 1993 for robberies and 1982 and 1994 for thefts. The vertical axis is the rate of recorded offences, set to be 100%, and the stages along the horizontal axis is the number of offenders, as a percentage of the offence figure. Such mixing of two units of count does not pose a problem provided that it is always kept in mind, and one category is not treated as if it were the other.

The first feature which stands out in all charts is the quite steep slope from offences to suspected offenders. In the UN questionnaire, that is defined as `the first formal contact with the criminal justice system', and in some sweeps is referred to as `apprehended'. Both terms have been used because this seems to be a concept with which some countries have difficulty. That steep slope represents, approximately, the proportion of offences not cleared up combined with the number of repeat offenders. As mentioned in the previous section, the `offences' base rate can be made high by the recording of every possible offence, so that while a steep slope might be created by a police force which was ineffective, it might also be created by a police force which was scrupulous in its record keeping. It is therefore probably not a good basis for comparison between countries, but it may well be a good basis for comparison within countries across crime types and/or over time.

The second feature requiring comment on the Norwegian charts is that the value for 1982 at which the first slope ends is very different, roughly 40% for robbery and 10% for theft That can be explained by at least three factors, which are not mutually exclusive; the explanation is probably a combination of them. First, "robbery" is rarely if ever a minor or trivial crime, so that the police will react to every instance of it. "Theft" on the other hand is both much more frequent and sometimes minor. The conscientious police will record it, perhaps for the insurance needs of the victim, but may take little action. Secondly, because of the higher seriousness and profile of robbery, the police will invest a lot more resources into identifying the offender. Thirdly, the population of known or potential robbers is much smaller than that of thieves, so that the task facing the police is better defined. However, for 1990 the ratio of apprehended to recorded offences is the same as for both years in Sweden; therefore presumably one or more the factors outlined above ceased to be the case.

The third noticeable feature is that in Norway, for both types of crime, the line for the progression through the system is almost flat, although Norway did not provide data on admissions to prison in 1990, so that the line is incomplete. The situation seems to be that, of those apprehended, very few are filtered out. While most people arrested for robbery may be expected to be sentenced to imprisonment, as in many countries quite a large proportion of those arrested for minor theft are given non-custodial sentences. The contrast with Sweden, to which we turn next, is noticeable.

The first contrast between the two countries is that, while the initial steep slope is much the same, and presumably for the same reasons, the difference between the values for the `suspected' category of "robbery" and "theft" is much less in the case of Sweden. The "theft" figure is lower even than for Norway, but not by much. The "robbery" value, though, is only half that of Norway for 1982, but about the same for 1990. The explanation for the earlier discrepancy, which could range from the fact that Sweden is a much more industrialized country to differences in law enforcement practices would have to be established by local enquiry, which might be of considerable interest to Scandinavian criminal justice authorities. One factor which it is important to note here is that, because all these charts are in percentages of a fixed, artificial base rate, they give no information as to the actual incidence of the events.

Both charts show a slight rise between `suspected' and `prosecuted', the effect being noticeable in the case of Sweden. That may well be a function of the way an apprehension is defined in local administrative law, or reflect a practice of bringing some offenders into the system at the prosecution stage as the first instance. The Swedish charts also show a clearer tendency not to send all those convicted to prison, especially those convicted of theft. The Swedish charts also show a greater proportion of those prosecuted for robbery not being convicted compared with the Norwegian chart. Again local study would be necessary for a definitive explanation, but the hypothesis of the effect of a much higher case load would be the first to be explored.

These examples from two similar countries illustrate the way in which the data in this set can provide a factual basis on which to build a set of questions which are focussed and practical, but which cannot be answered from the data set on its own. They might bring to the attention of policy makers in the countries concerned differences and similarities which had not previously been appreciated.

To illustrate the differential processing of adults and juveniles by the criminal justice system, and especially the different rates at which they are discharged from the system at the various agencies, four countries were chosen, two Asian and two European. As was explained in the preceding section, the breakdown by age was requested in the UN questionnaire only in totals, rather than by crime types, at each agency stage. Therefore each of these charts is for "all crimes", and there are probably significant differences in the ways that number is generated in different national offices. While, therefore, comparison of the raw numbers could be misleading, comparison by ratio avoids that particular set of problems, although the trade-off is that there can be no consideration of the size of the country or the size of the crime problem relative to the population total.

Charts 1 1 & 12 show the attrition rates for Finland and the Republic of Korea. In these two charts, the vertical axis is the first stage of the process, `suspected', and not `events' (recorded cases) as in the previous charts. There is no point in referring back to a total crime rate, and the resultant graph can be on a larger scale and so demonstrate changes at each stage more clearly. Both countries emerge as low users of imprisonment, although this characteristic tends to be the case for most countries when "all crimes" is the vertical axis, because most crimes are relatively trivial and rarely punished with imprisonment. The serious but statistically rarer crimes are overwhelmed and hidden in such an analysis. A detailed study of differential imprisonment rates between countries would have to be done by crime type. What is most noticeable about Finland is that the bulk of the diversion of juveniles is done at the first stage, that is, the case is not brought to prosecution; but if the prosecutors do decided to bring a formal case, it is almost certain to succeed. Presumably this reflects a policy of prosecuting juveniles formally only when it is a quite serious case which is not in dispute, and that in turn appears to be the case with adults also, but starting from a higher base. The graph for the Republic of Korea is included to demonstrate one feature which calls for some care in this type of analysis, namely that of a missing variable. As no data were given for the prosecution stage for juveniles, the graph becomes a straight line, from suspected to convicted, so that it is not possible to tell whether the profile for juveniles is in fact very close to that for adults. It does seem that the rate of imprisonment following conviction is slightly higher for juveniles than for adults in Korea, and that in itself might be thought to be a question worth further study

Charts 13 & 14 show the adult/juvenile filtering distinction in Italy and Japan. If the data are valid, the alternative scenarios are a remarkable contrast. The contrast is heightened by the fact that the attrition funnels in respect of adults are almost identical. If the two charts are overlaid on transparencies, it is almost impossible to distinguish between them in the case of adults. The situation for juveniles is the opposite for two thirds of the process, and then converging again in the last stage to give a very similar low rate of imprisonment. The reader is reminded again that these are charts of ratios, and not of actual numbers.

It seems that the Italian tradition is to locate the formal decision to divert or otherwise process juveniles who have been formally brought into contact with the criminal justice system with the prosecution stage, whereas the Japanese tradition is to divert their juveniles out of the system at the first possible opportunity. The Italian approach seems to incorporate some element of due process. On the other hand, the Japanese approach would both minimize stigma to the young offender and deliver an outcome quickly, thus avoiding delays in justice which are often regarded as undesirable as informal justice. Also the Japanese approach presumably would be much less costly in resources.

This last example illustrates again one of the main uses of this whole data set, which will be considered at greater length in the next section. That is the provision of empirical data as the foundation for international exchanges. If the data are valid, a meeting on the advantages and disadvantages of each of the approaches and the possible development of a method to obtain the best of both would have a basis of "fact" which could provide experts on both procedures with a foundation on which to build. One promising tool for the future of cross national research is the study of decision making, and where a decision is located in the justice process is an important first datum in such an approach.

The examples of filtering out from the criminal process by gender is illustrated by Charts] 5 & 16 for Finland and Japan. The Finnish chart shows that gender egalitarianism is practised by the agencies of criminal justice in that country, in that the two lines overlap completely until the final stage, at which point a smaller proportion of convicted women than men are sent to prison. The Japanese data, in contrast, show a clear tendency to divert more women than men from the process, but interestingly that occurs almost entirely at the first stage, the move from first formal contact to formal prosecution. Thereafter the proportion of women diverted is actually smaller than that for men; the explanation is presumably that the majority of those women who are left in the system after the large initial diversion are cases against whom the charge is both uncontested and quite serious.

The allocation of resources to the different agencies can be most clearly represented by a pie chart. Charts 17 to 21 inclusive give examples, two European and three Asian. These countries have been chosen partly because they are among the relatively few which provided complete data, and partly because they suggest, very provisionally, an interesting regional pattern which might repay further analysis. The methodology for reproducing such charts is given in Technical Annex 6.

The two European countries both show a country which, in terms of personnel, has somewhat over three quarters of its resources in the police, somewhat over one tenth in prison posts, and about one in twenty (Belgium) or one in ten (Hungary) in the trial and court stage. Comparison with the parallel figures for other European countries suggest that they are roughly similar. There is, perhaps, a "Western" model. The data for the USA made such pie charts difficult to construct, but the more important point, discussed more thoroughly under the title "Federated Countries" below, is that as most of these personnel are employed at the State and local level, such charts should be drawn for the separate States. They could be drawn for Federal employees, but such charts would neither tell the reader much about the organization of criminal justice in the USA nor be comparable with national figures from any other country.

The three charts from the Asian countries of the Republic of Korea, Japan and Singapore show a different pattern. In all three the proportion of police to the other three agencies combined is much higher, and the proportion of personnel in the prosecution and court functions is very small. The same is the case also for Japan and Singapore in respect of prison personnel, but Korea is closer to the European ratio. The most important observation to be made is that it is not possible to deduce from these data why this is so, and any evaluation of which is the "better" or "worse" system would be irresponsible and unsustainable. On the other hand, it is possible to put forward some hypotheses which could then be tested, by the gathering of further data, not all of which would be numerical.

The explanation might (or might not) be one or more of the following, as they are not mutually exclusive:

(i) the greater number of police in the Asian model enables strongly pro active strategies of crime prevention to be used, so that the actual level of crime is less. That could be checked by comparing rates for different kinds of crime, prorated per 100,000 of population;

(ii) the Asian police divert a larger proportion of the cases which come to their attention away from further formal processing. That could be checked by examining the statistics for this decision;

(iii) trial procedures in the Asian countries are conducted in some way differently from in the West, a way which is in one sense more "efficient" in that many fewer person hours per case are required. Research combining descriptive and quantitative data would be needed to test that proposition, but it is testable;

(iv) people in Asia who are charged with an offence have to wait longer for the case to come to trial, because the input process (the police) are proportionately larger, and the processing stage smaller. That could be tested by examining data on times awaiting trial in different countries;

(v) Asian countries make less use of prison for some (all?) crime types. That could be tested by examining data on sentencing in relevant Asian and European countries.

The possible explanations for the pattern which seems to be present in at least these data have been spelled out in some detail to illustrate one of the main uses of the data, which is the topic of the next section.

V: THE USES AND LIMITATIONS OF THE DATA

This section begins with explanations of two features of the report which may seem surprising, namely the absence of discussion of prorated figures for countries and the absence of federated countries from the examples given.

Prorated values, that is the raw numbers provided by countries converted into a figure which expresses the prevalence and incidence of that event per one hundred thousand (100,000) of population, are frequently used in cross national comparisons. In comparing the same category of figures in different countries, such as specific types of crime, or admissions to prison, prorating is the proper procedure because it removes the effect of the different sizes of national populations. Anyone using this data set for direct national comparisons would be justified in using prorating, and the fact that this report does not consider that type of analysis in no way implies a rejection of the method. The explanation of its non appearance is that it is not needed for the type of analysis considered here.

The quality of the data returned from federated countries seems to be probably lower than that from smaller, centrally governed countries of similar cultural and socio-economic status. That would not be surprising, because in some federated countries, at least, the management of criminal justice is performed at the State or Province level, if not even more locally, with a federal system in place to deal with a very limited number of special types of crime. The USA is the prime example, and most federated countries lie somewhere along the continuum between the USA and centrally governed countries. Most of the data requested in the UN Surveys is essentially of a management kind, that is they are the data collected by the administrators responsible for the day to day running of the system. The central federal government does not have any great need of the everyday data collected at the state or local level, nor is there usually any value for these bodies in reporting such data to the central government. Each state may have its own counting and recording methods, and these may not be identical.

Answering requests such as the UN Surveys is therefore intrinsically quite a lot more difficult and costly for federated countries, and this fact is probably a major component in any explanation for the possibly deficient data supplied. A detailed analysis which would establish whether the suggestion of less complete data from federated countries is generally valid or not has not yet been carried out, but will be attempted in follow up work. One objective of this report, therefore, is to contribute to thinking and discussion as to how this whole exercise, of the UN administration of the surveys and the use of the data set by many others in research, can be used to stimulate and assist federated countries in creating a centralized data set which is valuable to themselves.

After that preamble on the ways in which the data have not been used, a review of the main ways in which it seems appropriate to use the data is itself appropriate. The word "appropriate" used twice to give emphasis to the position that there are not `right' and `wrong' ways of using the data, but `better' and `worse'.

Inappropriate ways could be defined as those which rely upon the precise amount of the change in any one variable from year to year in the values; and that this is true especially of rates of recorded crime. The basis for that proposition is that all the variables which have been checked for reliability show a level of change which cannot be viewed as a good basis for the level of confidence normally considered acceptable in social science. In short there is just too much variation which could occur through errors in reporting and recording. It is not known if these data are invalid, but they might be. Conclusions based upon such data might be correct, and nothing in this analysis asserts that they are not, but we cannot be sure.

A second probably inappropriate use of the data is to test hypotheses. That is to some extent a function of the kinds of data collected, but again arises primarily over the uncertain quality of the data. The data set as a whole is unsuitable for the relatively precise and complex nature of the inferential statistical techniques used in such activities. It is a long standing principle in statistics that subtle or sophisticated techniques cannot be used on unreliable data to compensate for questionable data quality, and the level of confidence in the results cannot be higher than the limits imposed by the known quality of the data.

The first appropriate use of the data is to describe, by tables and/or diagrammatically, the dynamics of the main components in the criminal justice process, by crime type or (not `and') by age and gender across years. A criminal justice profile of a country consisting of some eighteen charts at five yearly intervals would show the main changes in input and the subsequent stages of processing, and the allocation of resources used.

Such a picture of the workings of a criminal justice system could have two applications, at least. These are:

(A) the identification of some points of apparent uncertainty or seeming dysfunction, or of interest for some other reason, of which further specific, focussed research could be directed. That is, the models generated could be used as a framework for deriving one component of a research program in a ministry of justice or parallel body;

(B) at the international level, the provision of background information to enable a meeting, such as a workshop on comparative policy analysis, to start from a knowledge base to which reference back can be made as required, rather than it being necessary for each delegation or participant to describe the position in their own country as a preamble. My experience in almost thirty years of such international meetings is that a lot of time is used unproductively in reaching a point where new thinking and exchange of new information can begin. On questions concerning the operations of criminal justice systems, the provision of the kind of information demonstrated in the preceding sections, for the relevant countries, years and variables would lead to a more focussed discussion in a much shorter time. By analogy, the same case could be made for its support use in parallel activities of exchange of experience and information, and joint activities in policy analysis between the individual states or provinces of a federated country.

One underlying question which seems frequently to come to the minds of policy and decision makers in criminal justice, is whether there is any value in knowing what is done elsewhere. One appropriate use of this material could be to support the argument that there is. Other people might be handling a given problem in a different way because, from some overall picture like this, they realise that the solution to their problem lies in persuading another agency to change its way of doing something. The advantage of simple graphic representations of "the big picture" is that they show, by different slices of a pie or different slopes on a graph, that the same agencies in different countries have different patterns of processing; such pictures can be the first step in lateral thinking. One impression that I have acquired over many years interest in and periodic visits to different parts of the criminal justice machinery is that agencies do not take much interest in how others operate or how changes in their own operations could help other agencies. Clearly that is not a universal truth; but material such as this can provide a foundation for what the international community calls, perhaps more in hope than accuracy, "international exchanges of information and experience".

Such an approach might therefore turn out to be a model which could be used for the comparison of criminal justice management and practices in the different States of a Federated country. Such analysis would not indicate whether one method or policy is better than another, and it would not imply any external or central control, even if it mostly carried out centrally in the first instance as a preliminary to the States undertaking it for themselves. What it would make feasible is a basis for comparing the way in which two (or more) different States organize their criminal justice systems, and from which, if those States wished, relative evaluations of efficiency or whatever parameter is of interest can be measured. It would be a useful, but non-prescriptive tool, in facilitating the process by which States learn from the experience of other States. What is valid and appropriate between States of a Federated country is, of course, equally so between separate sovereign states.

The second of the `most appropriate' uses of the data set returns to the topic of hypotheses. While the use of these data for hypothesis testing was described as questionable earlier in this section, the use of the data for hypothesis generating is both appropriate and potentially very fruitful. Some examples have been given in the text above, such as the possible explanations of the difference between Asian and European resource allocation, especially in respect of prosecution and courts, or the contrast between Italian and Japanese differential processing of adults and juveniles.

Some relatively simple hypotheses can be created to see whether a typology of countries based on the structure and dynamics of their criminal justice systems is feasible and meaningful. For instance:

(a) is it the case that certain countries of any main group which have a similar pattern in, say, the filtering out of juveniles, can be placed in the same category along other dimensions, or in other respects do they belong more with other countries?

(b) do countries which have a very high (or very low) ratio of prosecutors to police officers (or judges) know that this is the case? Is it a deliberate policy or simply a continuation of past, unexamined, practices? Can it be shown, either from the survey data or by information gathered within the country, that this has any discernible impact on case flows (for instance, time spent awaiting trial, on which data were not collected in the UN surveys), or proportion of cases carried through to the next stage?

(c) can some kind of mean score for various ratios between agencies be computed, (e.g. 30 police officers to 1 prosecutor to 2 judges to 5 prison staff), and would that be of assistance in policy analysis? Would it be feasible and useful to try to evaluate such data in terms not only of the most common structure, but the "optimal" one (and what would be the criteria for optimality)?

(d) what light can be thrown on the operations of criminal justice systems by an examination of any discrepancy of the resources profile generated from money resources and personnel resources data?

Three observations, at least, apply to such questions:

(i) most would require the collection and analysis of further data collected for the purpose, i.e. field work of some kind;

(ii) such questions could provide a factual basis for international exchanges of information and experience in various meetings which would make them much more focused and specific than seems often to have been the case;

(iii) selected topics from an inventory of such questions would provide an empirical platform for a programme of international research at an appropriate location.

VI: CONCLUSIONS

1. The underlying rationale of any analysis of this data set should be predicated on the approach sometimes referred to as the "progressive reduction of uncertainty". That is to say, first, that one of the main functions of many pieces of research is to create the base from which the next generation of more specific questions can be derived, asked and answered; and secondly that the more precisely a question can be asked, the greater the information value of the response. This approach conceptualizes research as moving along a continuum from some kind of primordial epistemological chaos, in which, for instance, the problem is not to test hypotheses but rather to identify them, to one of conceptual clarity where the objective is to obtain precise values for clearly specified factors or dimensions. The values may or may not be quantitative measurements, but they very probably contain more information if they are. The UN surveys are best seen as an activity toward the starting end of that continuum, so that better questions are a reasonable expectation, but any answers, in the sense of hypotheses strongly supported, should be regarded with some scepticism, if only because of the quality of the data.

2. The most productive use of the data set will not take the form of an attempt at a comprehensive analysis, because that would be very long, much of it very tentative, and most importantly of interest to only a limited range and number of readers. Rather it will take the form of a number, possibly quite large, of separate analyses by different individuals or groups concentrating on a subset of the data, for instance by geographical proximity or cultural similarity of countries and/or crime type and/or other parameters.

3. As a consequence of the first two, the immediate priority is to make the data set available to as wide a constituency of potential users as possible, and in a form which is easily manipulated, even by those without much statistical experience or equipment.