1A conviction leading to detention is often the result of many charges against the defendant. A quarter of detainees are convicted for two offences, another quarter for three or four offences and finally, around 15% for five offences or more. In the official conviction statistics, from which the raw data is often extracted by external users, this phenomenon of “convictions for multiple offences” is managed under a rule which leads to the selection of a single offence, with the others being ignored. A large number of the offences that led to conviction are thus simply concealed. The judges arrive at judgements that produce multiple convictions for reasons that need investigation or because the indicted behaviors have multiple dimensions. In these two cases, retaining only one of the total number of offences leads not only to the obscuring of part of the information (activities of the courts/charges), but also to the warping of the information.
2This article is an evaluation of the problems created by so-called “first rank” statistics, an approach to the selection of the principal offence. We also evaluate the interest in not applying it, through a statistical exploration of new types of objects. It is important to reflect on this because this type of method is at the root of many official publications that are used by researchers. Moreover, this type of practice is relevant not just to France, because since 1992, Canada has registered offences based on the most serious charge of the indictment (MSO: “most serious offence rule”, the Canadian equivalent of the “first rank” statistics in France– see the note entitled Summary of Historical Adjustments to Crime Data for Ontario, 1997-2000). Australia is also affeted by this phenomenon. Indeed, researchers analyzing the criminal specializations of convicted people have criticized first rank statistics as being a significant obstacle to their initiatives (Fisher, Ross, 2006).
- 1 The social and legal implications of the judicial categorization have already been examined in the (...)
3The main effect of statistics of the first rank is a smoothing of the distributions, producing a partially artificial homogeneity. Whether we analyze theft, drug trafficking, or even a murder as we shall see, an identical labelling does not necessarily correspond to acts that are empirically indistinct, and this especially so when we consider the crimes and offences of the youngest prisoners in relation to other age groups, which we will use as an example. As obvious and classical as this observation may appear, many analyses function with this “surface homogeneity” bias (Aubusson de Cavarlay 1996), and observe the variation of a given offence following age (for example) as the variation of the same homogeneous entity, without taking into account the diversity of situations that it covers. We will base our work on comparisons by age in order to show the ambivalence of the classification categories, particularly when they are used individually, detached somewhat from the combinations they form in the context of convictions for multiple offenses1.
4All statistics on offences lead to three possible levels of definition:
that of the offence itself,
that of the conviction which groups together these offences,
and that of the individual who might be found guilty of multiple convictions.
5From the criminal record we see that out of 933,048 punished offences in 2007, less than half (432,050) relate to convictions for single offences (which represent 70% of 614,000 listed convictions), the rest (500,098) belonging to the category of convictions for multiple offences (Les condamnations 2006, 9, 238).
- 2 The simple attempt is not taken into account by this rule. In the penal code, a person who attempts (...)
6At the first level (the offence) all statistics are relatively simple to establish, because we focus at the level of the unity of registration. When we are at the “condemnation” level, the question becomes one of knowing how to characterize a collection of offences in an unambiguous way. The strategy consists of retaining from among this group a single offence, called the “principal offence”, and the method of selection that is used is inspired directly by the tripartite classification of the penal code: When there is an accumulation of offences within the same conviction, the offence of reference is the first one cited within the most serious category (crimes, offences, contraventions) (Statistical annual of justice, 2007, 146)2.
- 3 To mitigate these problems, the statistical tables of the Minister of Justice show both first rank (...)
7The problem created by this method is clear: the first offence listed in the most serious category can appear in the first position in these computerized lists because it is the most “serious”, but also because it is chronologically the first to have been investigated, when it is not simply a case of random order (a possibility raised by Tournier, 1996). And even if “in the ideal case”, it is the relative seriousness of the offences that determines their order of appearance in the extract of the decree, we can legitimately question the criteria that allow for the creation of an objective hierarchy of the offences (these problems of hierarchy are mentioned by Derre, 2006 in relation to the Belgian case), especially those of a list of the size of Natinf (see Box 1 below). The criterion used can be that of the maximum incurred sentence: in this case the problem shifts, and we can expect many ex aequo because the sentences, to the extent that they deprive of liberty, are pecuniary, or are imposed in the form of interdictions or obligations, do not express a range of variation as wide as the offences labels themselves, when they are expressed on a quantifiable scale3.
Box 1: the NATINF table
The NATINF classification (NAture of the INFraction) was created in 1978. Its objective is to collect in a single database most of the offences present in different codes (penal code, road traffic code, consumption code, rural code, etc.), of which it systematically follows current changes. At November 30, 2007, the classification contains 11,364 offences in force, and 17,161 offences in total. Some 6,000 offences not in force have been abrogated (for example, the offence of vagabondage or night time as an aggravating circumstance for theft), or replaced by others because of modifications to the incriminating act. When a modification happens in one of the texts, without the incriminating act changing in nature, the NATINF number is kept and only the label is modified.
Each offence is defined by a code, a date of implementation, its nature (fine, offence or crime), the defining and repressing laws, and the punishments incurred. The label is added to it, they correspond to “short descriptions” (for ex. “Forgery: fraudulent alteration of the truth in a writing”), generally, of about ten words.
Incurred punishments have their own classification, in order of seriousness: deprivations of liberty, financial punishments, incapacity, decay, ban, bond, publication, closing of an establishment, confiscation. Those among them that are quantifiable are in terms of duration (quanta, amounts, numbers or time span), depending on the NATINF to which they are connected (also, they can be cumpulsory or optional).
By acting as a reference table in different computer systems and computer centres, the NATINF constrains the list of choices to the different categories of personnel in charge of entering the administrative data (clerks of the TAP, institutions, courts). Thus, it encourages a standardization of files. Since 1984, it has been the source of all legal statistics involving convictions.
But the NATINF is also a tool for the use of prosecutorial personnel, which allows them to conduct computerized research based on key-words or text references. The links lead to incurred punishments, and to different codes articles which define and repress the offence.
Aside from the jurisdictions, the national criminal record is connected to the index. The other services that use the NATINF are the treasury (recovery of fines and legal fees), and the services authorized to record offences (police, gendarmerie, labor inspectors, transport controllers…).
A separate database, the “file developed qualifications justice” (to this day, non-exhaustive: 733 categories corresponding to the offences with the most convictions in the last three years, product of the old file COPJ) gives the “developed qualifications” (in our example: “to have to * , the * , altered, by whatever means, fraudulently the truth in *(characterize the facts), in a document, or all other means of expressing thought aimed at proving a right or fact having legal consequences, in the case in point * (type of document or support), to the prejudice of * (victim’s given name, family name)”).
8A problem even more disturbing can appear, because the sentence imposed for a given crime can be less than one imposed for an offence (see also Derre, 2006): the statistical rule of selection (the crime prevails over the offence which prevails over the minor offence) is thus at fault, because it leads to the selection of a given conviction of an infraction of a lesser gravity than those which have been ignored.
- 4 National Index of Prisoners. It is the compilation, created daily at midnight of management databas (...)
9Regarding the prisoner population, the problem is more difficult, because a prisoner can be on the index we are using here (the FND4) for many convictions, which themselves can involve many offences. We can thus be faced with the choice of establishing statistics either at the level of the conviction, with the problem we have just reviewed, or at the level of the individual, with a doubling of these because the reduction has to occur a second time.
- 5 The Nataff (nature of the case) is essentially of a penal nature. Its implementation dates from 199 (...)
10A solution which bypasses the problem is to use a list that is less refined than the Natinf, in the hope that including these items in a smaller number of broad categories will ultimately allow us to reorganize the multiple offences of an individual or of a conviction in one and the same entry. But this method implies a reduction in the original information, and does not prove to be completely effective as shown by the transfer from the Natinf towards the Nataff, an interlinked classification of a lesser breadth than the first5.
- 6 We retain here only one case per individual, going by the functional specifications of the FND.
- 7 Nataff 3 and the Natinf are linked by a table of correspondences which creates an unambiguous conne (...)
11The Natinf is the level with the most detail, recording more than 11,300 different offences. The Nataff is an index of the interlinked type. Table 1 thus gives us an idea of the quantitative scope of these problems, on the basis of two lists, the Natinf on the one hand, where the description unit is the type of offence, and the Nataff on the other hand, the interlinked three levels classification of lesser scale. We read in this table the number of different records by individual, according to the chosen precision level6. These four levels of details are defined on the one hand by the Natinf, and on the other hand by the different levels of the Nataff. At its finest level (Nataff 3), it includes 350 labels, then 85 for the intermediate level (Nataff 2), and finally, 12 at the most aggregated level (Nataff 1)7. The table shows that, whatever level we are on, the percentage of the individuals found guilty of multiple offences is important: 31% at the level of 12 large categories of the Nataff 1, then 44% and 59% for levels 2 and 3. Finally, at the finest level, 62% of prisoners appeared in the FND file with many offences attributed to them.
12The prison population displays a structure quite different from what we see for the total of convictions on the criminal record, of which 70% are single offence (cf. supra). Here only 38% of the convictions are simple offence (the impact of long sentences cannot by itself justify this difference). Prison gathers together more serious matters, for which the instructions are more detailed. We can thus see the special intensity of problems generated by the specific context of detention.
Table 1: number of offences by individual for different classification levels
Field: Detained at 01.12.2007
Interpretation: out of 65 055 detained at 01.12.2007, 15 003 committed two offences defined with the Natinf classification index.
13In the hypothesis with which we are trying to create some sort of statistic relating to conviction, that is to say, using the rule of selection described above (when there is an accumulation of offences for the same conviction, the main offence is the first one cited in the most serious category), we would have the following situation. Out of 65,055 prisoners, there are 31,015 (47.7%) of the records for which the selection of a “principal offence” does not present ambiguity: because it is the only offence of the individual or, if that is not the case, because it is the only offence in the most serious category (crime/offence/minor offence). For the rest, meaning more than half of the people imprisoned, that must be decided with the rank (see Table 2). To all of this is added the problem mentioned supra, the possibility for a minor offence to be given a sentence more serious than a crime, a scenario rendering the rule of selection still more counter-productive.
Table 2: number of concurrent offences for the status of principal offence (by individual)
Field: detained at 01.12.2007
(*): a single matter is retained by individual.
14All in all, this creates a considerable mass of individuals and offences for which statistics should be computed; statistics which, following the situation, would sometimes respect a chronological order, sometimes an order based on seriousness, or no order at all. Concerning the most serious convictions, those that lead to imprisonment, we see that the application of standard rules for selecting the main offence presents a still greater problem than for the totality of convictions.
- 8 In order to see what happens when we resort to this mode of selection, we put forward a method of a (...)
15There is another problem posed by statistics of the first rank that we are not exploring here: identical convictions, made up of the same series of offences competing for the status of main offence, will not always produce the same “winner” depending on the social or political contexts. The aggregated statistics are impacted and subjected to an effect of circularity similar to that shown to be involved in studies of social mobility (Thevenot, 1990, 1994): indeed, in such a case, the first rank will tend to produce, at the aggregated level, statistical distributions conforming to the logic of selection that is characteristic of a lesser level, even while the individual offences making up the multiple convictions remain identical8.
16It thus seems imperative to consider these individual records, without continuing to base things on the statistics of rank one, which causes us to lose too much information. The rest of the paper proposes several statistical possibilities which show the benefit of analyzing the phenomenon in a more direct way.
17The substance of our argument in this second section rests on this fact: the offences studied in the first part do not, as we have seen, correspond to a simple equation “one prisoner=one offence”. Regarding the prison population, it is rare that a single offence is held against a convicted person. Depending on the level we focus on (natinf, or nataff of level 1, 2 or 3), the proportion of individuals having just one record in the file indeed varies from 38% only at the most refined level (natinf), to about 70% at the most aggregated level (the nataff of level 1. which includes only 12 large categories; cf. Table 1).
18In these conditions, it is possible to work on these groupings of individual offences by considering what they gather and what they do not gather. For that we can consider that every grouping of crimes or offences in the same conviction is a source of so many mutually uniting links. With a population of 65,000 prisoners (for 166,000 offences-natinf), we thus create a relational system which includes, at the most refined level (natinf), a collection of 455,000 associations of crimes and/or offences associated in the same convictions (330,000 for the nataff 3, 134,000 for the nataff 2, 63,000 for the nataff 1).
19Table 3 gives us an outline of one of the measures that is usable to understand this relational system, with the example of murder. It shows the odds of such and such a type of offence being associated with murder for the same condemned person, always relative to the state of these associations in the age category of 18-25 years (it is taken as a “reference”, thus explaining the presence of “1s” only in the corresponding column). The offences related to murder are thus classified from top to bottom, from those most specifically associated with homicide in the 18-25 year olds, to those who are the least.
20Let us first look at the odds that indicate associations more familiar to youths convicted for murder. Theft, possession of drugs, misbehavior and acts of violence are the offences most specifically associated with them. The possession of drugs, for example: those over 45 years of age are about 10 times less likely to be guilty than those between 18-25 years (odds of 0.1, in the last column of of Table 3). If we look at all the lines of the table for which the odds are less than 1 for the other age groups, we can then develop a first idea of the specificity of murder in the youngest convicted relative to the others. It is characterized by a more frequent association to theft, drugs and to acts of violence.
21We now come to the last lines of the table. Here, the odds are inverted: it is offences associated with murder that are more heavily associated with older convicted people. Firstly, we find homicide, the category to which murder itself belongs: it indicates, besides murder, another member of its category, in 76% of the cases, assassination. It is therefore mostly premeditation that differentiates the oldest of those convicted for murder in relation to the youngest: a 50% rise in the odds of association among the 25-45 year olds compared to the 18-25 year olds, then, 70% among the older than 35 years. Then comes rape: on average twice as much chance within the oldest group (always compared to the 18-25 year olds). Violence between spouses comes last as a type of offence most specifically connected to murder for the oldest.
- 9 All this depends, at a minimum, on a double filter: on the one hand, that of the offences registere (...)
22If we summarize these observations, it can be seen that murder among the youngest of the convicted people occurs more often than with the other groups, accompanied by one or many thefts, acts of violence and possession of drugs. With the oldest, it is more often premeditated (assassination), tied to a sexual assault9, or to acts of domestic violence.
Table 3: offences associated with murder according to the age (in% and ratio)
Interpretation: the first line (18-25 years) is used as a reference. The numbers from the other lines represent the chances of seeing an offence linked to the offence of murder depending on the age group, always relative to those 18-25 years old. For example “0,4” signifies that inmates older than 45 year incarcerated for murder have around half the chances of the 18-25 year olds of also being convicted for “associating with criminals”. Inversely, “1,7” signifies that those older than 45 years accused of murder have close to twice the chances of also being accused of rape as the 18-25 year olds.
23In all of this, we emphasize the methodological aspects: when we speak of “murder”, for example, even if it involves an homogeneous label in the penal code or in the official statistics, we are not referring to the same thing, depending on the age of the condemned persons. Characterizing a conviction for multiple offences in terms of only one of those offences, as detailed as the list we choose the label from may be, reduces the diversity of the object, like any task of codification does.
24Statistics computed based upon this first rank method reinforce this reduction, by choosing from an object that has already been fragmented into apparently countable parts, a single element, the most important, with the limits that we underlined in the first part of this paper. On the contrary, what the offences statistics need, is to take into account the full extent of these phenomena by using a measuring scheme that is less dependant on the employed rule. Constructing complex offences through sets of combinations, in opposition to the simple offences actually used, can be done by resorting to procedures that are simple enough to implement, and that would increase the correspondance between the labels used and the realities they are supposed to describe.
25Still on the new methodologies that could be used when dealing with convictions for multiple offences, the use of this combinatorial method can also be made systematic (rather than applied to a given topic, in our example, murder). We can indeed choose to no longer base our work on a simple example like we have done here, but rather to consider the corpus of offences that is not limited to a single case. In this situation, the results can’t be displayed in a single table, since one has to consider a system of associations among offences that reaches several thousands of them and five to ten times more associations directed toward the same object. As a comparison, Table 3 gives an example, in which information dicussed represents about 1/100,00th of this structure.
26Can we see this structure? Not at the level of the natinf with its 11,300 labels, the result would be unreadable. However, each record in the natinf is associated with another interlinked label: the nataff. This one contains around 350 labels at its most refined level, 85 at its intermediate level, and 12 large categories at its most aggregated level.
27The following graph shows the relationships between the type of offence, at an intermediate level of detail (Nataff level 3, about 350 labels). For greater readability, the names have been replaced by numbers. We see here only a part of the types of cases that exist in the categorization (68 here); these are interconnected by more or less dense lines depending on the degree of statistical association between the two cases. The selection of the units is made based on their representativeness within the data (we use only the most important data for this example, but any scale is possible). These are links based on the order of their appearance (a given type of cases is more or less linked to others, here we select only the first two links).
Graph 1: network of first statistical links between principal type of offences (nataff 3)
Interpretation: each node represents a type of offence. Each link shows an important statistical relationship between two nodes. The link is more or less thick depending on the strength of the relationship. Each relationship is linked to the simultaneous presence of two types of given offences in the same conviction.
28If we read this graph, we first see clusters that are more or less isolated from each other. Let us read clockwise.
At the top right is all that relates to highway offences (17: (not having driving licence, 14: driving under the influence of alcohol or drug, 58: violation, restriction concerning the right to drive, 3: lethal traffic accident, etc.). This cluster is not isolated from the whole, but bound by a particular type of case, an outrage against the justice authority / obstacle (45).
Then, in the lower right corner, we observe a cluster of very dense cases: drugs (22, 54, 53, etc.), where possession, carrying and use are well connected. Added to this hard core is money laundering (12), pimping (46), criminal conspiracy (8), customs offences (34).
To its left is a third block to which it is linked by the legislation/possession of weapons (37): this block is made up of acts of violence (62: violence with or without ITT (Incapacité Totale de Travail-rendered totally incapable of working), 60: violence between spouses, 61: violence towards public authority, 21: destruction or degradation of public property, 41: threats, blackmail, 44: insult to a public agent).
Linked to this third block by the intermediary of violence with ITT (59) and the destruction/degradation of public property is the largest block on the graph: that of theft, receiving stolen property, swindles, breaches of trust, frauds etc. Within this block we also find voluntary homicide (32), kidnapping and illegal detention (23).
We finish with the small group of cases found at the bottom left of the graph, of which we will guess by elimination that it involves sexual abuses (6: sexual assault of a minor; 15 corruption of a minor; 38: ill treatment; 56: rape of an adult; 57: rape of a minor). This time, there is no link connecting it to the whole, probably because of its special nature.
29Without going further on this short tour around the display of types of offences and their associations within multiple convictions, we observe, as we would for any classification, the large families of deviances leading to prison and the links that unite them. Compared to one classification, two advantages: that of seeing the “bridges” that unite these families in a single “network”; that of being able to observe the core of these families (17: absence of a driving permit; 22: possession of drugs; 62; violence with or without ITT; 10: other aggravated thefts; 68: simple theft; 26: swindle).
- 10 This operation is made by looking at cases that are at one or two steps from the selected cases, no (...)
30The graph which we have just outlined shows relationships that are more or less obvious. We must now ask ourselves what will become of this network if we no longer observe it regarding everyone but just in terms of prisoners’ ages. Graphs 2 and 3 are more detailed than the one we have just seen (we widen the selection of units), only this time, they are centered around one particular type of case: the voluntary physical assault of an adult10. Graph 2 relates to those under 30, the following represents those over 40 years old.
31In Graph 2 (less than 30 years), we see quite a significant degree of dispersion of associations (many dense lines) around the type of offence studied. We don’t want in general terms to describe the offences linked to this age group. On the contrary, we are targetting one particular category of offences in order to show what gravitates “around” in this age group. The offences associated with the physical assault of an adult thus inform us regarding the particular nature of this type of case, in the relevant age group. The associations are mainly with destruction/degradation, offences against public authority, offences against the dignity of the person, different types of thefts, and offences related to driving capacity.
Graph 2: network of main relationships between cases related to phyical assaults upon an adult for prisoners of less than 30 years at 01.12.2007
Interpretation: cf. graph 1.
32Let us now observe what is occurring with those over 40 years of age. Here, the dispersion around the targetted offence is very weak (few dense lines), and the hard core is made up of the offences against mores and the corporal offences on minors. Behind this core we recognize sexual offences (cf. Graph 1), with a link to voluntary homicide, from the second rank but nevertheless important.
33We can thus conclude that for the same case (corporal attack upon an adult), statistically, we do not face the same acts depending on the age of the prisoner. The comparison of two graphs thus shows the hidden heterogeneity of an apparently “homogeneous” label. This reinforces the conclusions that we reached from Table 3. The study of offences, when done individually, partly conceal the nature of its object. The multiple offences, analyzed in this manner, can help us to reconstitute it.
Graph 3: network of main relationships between cases related to physical assaults upon adults associated with prisoners older than 40 years at 01.12.2007
Interpretation: cf. graph 1.
34What is interesting about this type of method is that it aims at a reduced distortion of the object described. The classification used is detailed, and it is based on elements that are marginally identified: its use creates a segmentation of acts that are subjected to judgement, thus the convictions for multiple offences are represented significantly within the prison population. The use of a rank one method appears incompatible in such a situation, because it does not take this complexity into account.
35It is important to not neglect the information that these convictions convey, as much about the practices of judges as about the deviant behaviours concerned, by reducing them as it is done while referring to statistics of rank one. These have significant limits that are avoided when we use all of the information linked to a conviction. In the second part we saw that considering the relationships between offences that produce multiple convictions creates new and interesting possibilities.
36In spite of all these conclusions, considering and analysing multiple offences is rarely done. When it does occur, it is most often in the form of aggregated numbers shown with the body of registered offences, and with the main offence, that allows us to find the identification unit, individual, conviction, or sentence (with important differences following, for example, the ILSs (Infractions à la législation sur les stupéfiants - Drugs related offences) are hidden and play a very minor role in the statistics of rank one (see especially Barré, 2008), while they are very present at the level of the PPLs (peines privatives de liberté – Prison sentences). It is possible however, by creating a combined classification, to keep the advantage of a formulation that is exhaustive, yet still readable. For the rest, more complex relational methods allow for the analysis of original objects that we have tried to briefly present.
37In its traditional usage, the Natinf tends to have a double function, simultaneously serving to standardize the treatment of cases (automation with regard to forms and mailings), as well as in centralized services of administration, establishing in turn its statistics and annual reports. This is also usable upstream because it takes the form of a search engine using key words that allow, for example, to research which contravention of the 5th class to choose for this or that type of offence, or to reach the list of incurred punishments corresponding to offences eligible for a given case. It is therefore a tool that fits in with the new imperatives of productivity within the justice system (cf. especially Bastard, Mouhanna, 2007, 59, 67, 72, 75, 78, 103, 120-130), and more generally with the implementation of quantified procedures for evaluating public service effectiveness (Desrosières, 2004).
38The parallel between texts of laws and statistical classification, the obligation to use the latter within legal settings, and the availability of computerized research tools all combine to make penal identification and statistical codification two acts that are quasi identical and simultaneous. They are two different working logics which determine a same object, and in this particular case we must expect the legal logic to dominate the statistical one because the “conventions of equivalence” (Desrosières, 2001) are here produced by the penal code, whereas the coding is operated by the professional of the justice system. However, the development of an index like that of the “developed categorizations” (cf. the box 1 above – and more widely, the existence of scales mentioned in Bastard, Mouhanna, 2007) poses the question of the consequences of this rationalization: in return, can the statistical logic penetrate and reinforce the effects of circularity already alluded to above (Thévenot, 1990, 1994)? The problem has to be confronted with a more detailed fieldwork, but it is clear that the statistics of offences (with or without adequately considering multiple offences) remain dependent on this type of mechanism.
39At the end of the process, the published figures obscure the two important stages of the statistical production line. The first is the systematic recording of offences. It is closely tied to the rationalization, by an interposed computerized tool, of the activity of the jurisdictions. The second consists of the publication and the availability of aggregated figures that are ready for use which, by existing, considerably reduce the complexity of the cases which they aim to document. It is important that these different stages be known by the user. The first one in particular represents an important axis of an endogenous character (Desrosières, 2000; Blum, Goussef, 2003 give a good example) of these statistics, whose form can be affected by the contemporary orientations of the rationalization and acceleration policy in the treatment of cases processed by the jurisdictions. The result of the application of a first rank method then again contributes to the transformation of this object in a manner that is out of the control of the user, who does incidentally, send the case into a “second life” (Desrosières 2004). It is then detached from the complex gestation whose impacts are little known, a process which precedes its objectivization as official statistics.