Australian Institute of Criminology

Skip to content


In this section, an overview of the longitudinal offender cohort that was used in this project will be provided. The five phases involved in the research will then be outlined. First, the process used to establish the offender cohort will be examined. Second, the analytical strategy adopted to assess the number of offender trajectories and their characteristics will be described. Third, the costing approach that was used to assess the cost of individuals in the different offender trajectories will be outlined. Next, the approach that was used to assess whether some communities were more likely to generate chronic offenders and to explore the extent of residential mobility will be reported. Finally, the approach that was adopted to determine whether the communities that generated the most costly chronic offenders could be identified will be outlined.

Longitudinal offender cohort

The longitudinal offender cohort consisted of all individuals born in 1990 who committed an offence (other than traffic and breach offences) in Queensland and were formally cautioned, referred by police to a youth justice conference, had a finalised youth court appearance, or had a finalised adult court appearance when aged 10 to 20 years old. There were 14,171 individuals in the final research sample, of which 9,949 (70.2%) were male and 1,895 (13.4%) were identified as Indigenous. The average age of offending onset was 16.21 years (SD=2.38). These individuals were responsible for 71,413 offences. Most offences committed by cohort members were property or public order related (see Table 3).

Table 3 Offences committed by cohort members
Offence types n %a
Theft and related offences 20,651 28.9
Unlawful entry with intent/burglary, break and enter 10,585 14.8
Public order offences 10,479 14.7
Property damage and environmental pollution 8,069 11.3
Offences against justice procedures, government security and government operations (excluding breaches) 5,763 8.1
Illicit drug offences 4,870 6.8
Acts intended to cause injury 3,567 5.0
Dangerous or negligent acts endangering persons 2,051 2.9
Deception and related offences 1,984 2.8
Miscellaneous offences 1,139 1.6
Weapons and explosives offences 863 1.2
Sexual assault and related offences 638 0.9
Robbery, extortion and related offences 553 0.8
Abduction and related offences 194 0.3
Homicide and related offences 7 0.0
Total 71,413  

a: Percentages may not total 100 due to rounding

For these offences, individuals had 33,455 criminal justice system events (see Table 4). A criminal justice system event involves a caution or police-referred conference taking place or a finalised youth/adult court appearance. Of the 14,171 individuals, 7,215 had at least one caution, 824 had at least one police-referred conference, 2,337 had at least one finalised youth court appearance and 12,097 had at least one finalised adult court appearance. About one-third (34.5%) of individuals only had contact with the youth justice system, with two-fifths (43.2%) only having contact with the adult system and one-fifth (22.3%) having contact with both the youth and adult systems (see Table 5).

Table 4 Criminal justice system events involving the cohort (n)
Event type Events Distinct individuals a
Caution 9,799 7,198
Police referred conference 984 822
Children’s court appearance (finalised)b 6,199 2,130
Magistrates court appearance (finalised) 15,959 9,201
District court appearance (finalised) 471 433
Supreme court appearance (finalised) 43 42
Total events 33,455 14,171

a: Distinct individuals may have had more than one of each event type

b: Children’s court includes Children’s Court and Children’s Court of Queensland

Table 5 Individuals in cohort who had different types of events (n)
Event type n %a
Caution only 3,799 26.8
Youth Justice Conference only 104 0.7
Youth Court only 436 3.1
Adult Court only 6,123 43.2
Caution and Youth Justice Conference 150 1.1
Caution, Youth Justice Conference and Youth Court 78 0.6
Caution, Youth Justice Conference, Youth Court and Adult Court 261 1.8
Caution, Youth Justice Conference and Adult Court 140 1.0
Caution and Youth Court 307 2.2
Caution, Youth Court and Adult Court 800 5.7
Caution and Adult Court 1,663 11.7
Youth Justice Conference and Youth Court 14 0.1
Youth Justice Conference, Youth Court and Adult Court 23 0.2
Youth Justice Conference and Adult Court 52 0.4
Youth Court and Adult Court 221 1.6
Total 14,171  

a: Percentages may not total 100 due to rounding

Research phases

Phase one: Establishing the offender cohort

The offender cohort was created by linking between the cautioning dataset (Queensland Police Service), police referred conferencing dataset (Queensland Police Service), youth court dataset (Department of Communities) and adult court dataset (Department of Justice and Attorney General). The process used has been described elsewhere (Allard et al. 2009), but involved three steps:

Agencies provided identifying information (but not case information) to the Office of Economic and Statistical Research (Queensland Treasury) and case information (but not identifying information) to Griffith University. These datasets included agency identification numbers that were used to link between the identifying and case information datasets.

Within the Office of Economic and Statistical Research, a researcher linked within and between the datasets based on identifying information, including name, surname, date of birth and sex. Each unique person was assigned a Griffith University identification code. Agency and Griffith University identification codes were then released to Griffith University. Griffith University identification codes were assigned to the case information to identify distinct individuals for the purposes of analyses.

After linking, there were 90,785 offences finalised across systems, involving 16,558 distinct individuals. The data were cleaned to resolve inconsistencies between systems in the core demographic variables of age, sex and Indigenous status, and missing values were propagated from the known values in another record based on the balance of probabilities. After resolving discrepancies, sex was missing for 11 (0.1%) individuals and Indigenous status was missing for 1,217 (7.4%) individuals. All missing data for sex related to contacts that individuals had with the adult court system. Most individuals who did not have an assigned Indigenous status were from either the cautioning dataset or the adult court dataset. Individuals who were not identified as Indigenous were assumed to be non-Indigenous.

Given that an offender cohort was being created, all offences that resulted in a not guilty (n=1,445) finding were excluded because they did not represent offending. Two offence types were also excluded from the dataset. Traffic and related offences (n=15,077) were excluded because most are dealt with by Infringement Notice and individuals can elect to have a court hearing. Breaches of court orders (n=2,850) were excluded because they may not represent additional offending. After these exclusions, there were 71,413 offences committed by 14,171 offenders.

Phase two: Exploring the number of trajectory groups and their characteristics

A dataset was created to address the first research question How many distinct offender trajectories can be identified? The dataset had the annual number of offences for each of the 14,171 offenders in the cohort based on their age at the time of offence. To calculate age at time of offence, the individual’s date of birth and the earliest recorded date for each offence were used because the actual date of offence was not recorded. For cautioning and conferencing data, the date of offence was usually the date when the offence was reported to police. For court matters, the earliest date was either the date of lodgement or the earliest court appearance relating to the matter.

Nagin and Land’s (1993) SPGM was used to model offence frequency annually over the life course when individuals were aged 10 to 20 years old. The SPGM analysis was undertaken using the SAS procedure ‘PROC TRAJ’ developed by Jones, Nagin and Roeder (2001). As the majority of individuals in the cohort offended for short periods of time, there was an excess of data cells with zero counts for offending. Because of this, the offending count data was distributed according to the Zero-Inflated Poisson distribution (Fergusson, Horwood & Nagan 2000; Nagin 1999). Additionally, several individuals had high annual offence counts that exceeded 20 offences in a given year (n=279; 2%). These outliers were scaled to enable the trajectory analysis to converge.

Given the non-parametric nature of the procedure being used, it was necessary to specify the number of trajectory groups being modelled and their form prior to analysis. Thus, the development of the final model was necessarily iterative, with the process being repeated a number of times to determine the parameters that produced the best fit for the data. The final number of trajectories for the model was determined based on both the Bayesian Information Criterion (BIC) and the average probability of group assignment. The BIC increases as the model fit improves (incorporating the penalty for increases in the number of trajectories), while the average probability of assignment is higher for models with more distinct trajectories (Nagin 1999; Piquero 2008). Thus, the model with the optimum number of trajectories needed to have a high BIC (relative to other model options) and an average probability of group membership that was as close to one as possible.

The trajectory group membership that was assigned to individuals was then linked to case information to explore the second research question What are the demographic, offence, and criminal justice system event characteristics associated with trajectory group membership? Demographic characteristics examined included sex and Indigenous status. The types of offences committed by individuals in each trajectory group were explored. Criminal justice system event characteristics examined included type of event and number of days sentenced to community based supervision and detention/incarceration.

Phase three: Assessing the costs of offender trajectories

Two approaches were used to address the third research question What are the costs of offender trajectories? Criminal justice system costs of individuals in the trajectory groups were assigned based on the interactions they had with the criminal justice system, while wider social and economic costs of crime were assigned by updating Rollings’ (2008) assessment and applying costs based on offence type.

Criminal justice system costs

Criminal justice system costs were estimated based on the costs of criminal justice system events and supervision costs. These were assessed using the Transactional and Institutional Cost Analysis (Carey, Waller & Marchand 2006). This approach views offenders as consuming resources when they have transactions with, and are processed through, the criminal justice system. One strength of this approach is that it enables an assessment to be made about the cost of resources invested by multiple agencies. Although the Transactional and Institutional Cost Analysis is frequently used to assess costs at the micro level, the approach was used to determine the average cost of practices as individuals flowed through the criminal justice system.

Figure 1 presents a schematic diagram of the transactions individuals have as they flow through the criminal justice system. The average cost of police, court and supervision practices were assessed for youth and adults. Average police costs were calculated based on publicly available information and an internal police time-in-motion study, which assessed how long particular practices took for youth and adults. Five steps were used to assess the cost of police responses:

  • 35 percent of the 2010–11 police budget was directed towards crime management ($624,796,550; QPS 2011a, 2009).
  • Examination of police practices indicated that 9.3 percent of offences were dealt with by ‘other’ and this proportion was subtracted from the crime management budget (leaving $566,440,552; QPS 2011b).
  • The number of youths and adults cautioned, conferenced and processed through the courts during 2010–11 were examined and total hours was calculated based on how long practices took in the Queensland Police Service time-in-motion study (DJAG 2011a, 2011b; QPS 2011b, 2005).
  • The average hourly rate was assessed as $245.1, calculated by dividing the remaining crime management budget ($566,440,552) by the total time police spent processing offenders (2,311,118 hours).
  • The cost per event was calculated by multiplying the length of time that processes took police by the hourly rate.

Average costs per court finalisation in the Childrens, Magistrates, District and Supreme courts were based on figures provided in the Report on Government Services (Productivity Commission 2012). The average cost of youth conferencing was determined by dividing the overall youth conferencing operating budget ($9.3m) by the number of referrals (n=2,614; Department of Communities 2009). The cost of community-based supervision and detention for youth was assessed based on the most recent costing information which was available (Bleijie 2012; CAIR 2008), while these costs were assessed for adults using costs provided in the Report on Government Services (Productivity Commission 2012).

Figure 1 presents average costs for the main transactions that individuals had with the criminal justice system. Transaction costs were added to calculate the cost per finalisation. For example, police cautioning only involves police expenditure (either $1,275 per youth or $1,103 per adult). However, the cost of individuals appearing in court requires police expenditure ($3,701 per youth or $2,696 per adult), court expenditure (depending on the level of the court) and possibly supervision costs, which were assessed per day.

As information was only available about the number of days that individuals were sentenced to various forms of supervision, it was assumed that youth would serve 60 percent of their detention sentence, while adults would serve 80 percent of their incarceration sentence before being released. These assumptions were based on advice provided by the relevant agencies about the applicable average proportions that would be subject to early release. Consistent with practice in Queensland, individuals were assumed to serve 100 percent of time sentenced to community-based orders. Where more than one court outcome was recorded at an event because several offences were finalised, it was assumed that sentences would be served concurrently and the most serious outcome for the event was used.

Wider economic and social costs

Estimating the wider economic and social costs of crime is challenging and there is considerable variability in these costs depending on whether a bottom-up or top-down approach is used. While bottom-up approaches include a range of specified tangible and intangible costs, they result in lower estimates than top-down approaches (ie willingness to pay). Given the absence of published estimates based on willingness to pay in the Australian context, a bottom-up approach was used that involved updating an assessment about what these costs were in Australia during 2005. Rollings (2008) estimated the average economic and social costs of crime for 12 offence categories. These costs included medical costs, costs of property loss or damage, costs of lost output and intangible costs. Costs that were excluded from the study were justice system costs, costs related to providing government services to victims, and security industry and insurance administration costs. The study acknowledged that there was likely to be considerable variation in costs within each offence category, so offence characteristics were taken into account when assessing costs. For example, most offences against the person involved assessing the number that would have resulted in injury requiring medical treatment or hospitalisation. Property offences were assessed separately for residential and commercial offences and took into account the number of offences that resulted in insurance claims.

Table 6 presents the social and economic costs of crime based on an update of Rollings’ (2008) assessment. In mapping the costs from the original assessment to the Australian Standard Offence Classification (ASOC), assault was mapped to two ASOC categories Acts intended to cause injury and Dangerous and negligent acts intended to cause injury. Six offence types in the original assessment were subsumed by two other ASOC codes—Theft and related offences included four theft types and Property damage and environmental pollution included criminal damage and arson. Where more than one offence category in the original assessment was included in one ASOC offence code, average costs for the offence code were calculated. Average costs were based on ratios developed to account for the frequency of each offence category in Queensland during 2010–11 (QPS 2011b). The 2005 cost of each offence was then adjusted for inflation to determine the 2012 cost (RateInflation 2011).

Unfortunately, the average cost per offence type was not assessed by Rollings (2008) for six ASOC categories. Offences that were not costed include:

  • public order offences (n=10,479; 14.7%);
  • illicit drug offences (n=4,870, 6.8%);
  • offences against justice procedures, government security and government operations (n=5,763; 8.1%);
  • miscellaneous offences (n=1,139; 1.6%);
  • weapons and explosives offences (n=863; 1.2%); and
  • abduction and related offences (n=194; 0.3%).

Therefore, costs for these offences were not able to be included in the projected costs for the offender trajectories discussed in this report.

Table 6 Mapping cost of offences from Rollings’ assessment to ASOC
2005 assessment in Australia ASOC

Cost per offence

2005 ($) 2012 ($)
Homicide Homicide and related offences 1,915,323 2,293,376
Sexual assault Sexual assault and related offences 7,500 8,980
Assault Acts intended to cause injury
Dangerous or negligent acts endangering persons
1,695 2,030
Robbery Robbery, extortion and related offences 2,300 2,754
Burglary Unlawful entry with intent 2,869 3,435
Theft of vehicles (n=4,095) Theft and related offences 1,241 1,486
Thefts from vehicles (n=4,949)      
Shop theft (n=14,453)      
Other theft (n=7,563)      
Fraud Fraud, deception and related offences 21,370 25,588
Criminal damage (n=12,565) Property damage and environmental pollution 3,357 4,020
Arson (n=232)      

Source: Rollings 2008

Phase four: Exploring whether some communities generated chronic offenders and their residential mobility

Given that chronic offenders are likely to commit a high number of offences and be costly, the fourth research question was Are some communities more likely to generate chronic offenders than others? To explore this question, the proportion of the population in each POA who were chronic offenders was explored, based on each offenders first recorded residential POA. Chronic offenders included individuals in the moderate and two chronic offender trajectory groups who had a higher level of contact with the criminal justice system and committed more offences than members of the two low trajectory groups. From the trajectory analysis, 2,234 offenders were classified as chronic (as described in Characteristics of Offender Trajectory Groups in the Results section). Chronic offenders represented 15.8 percent of offenders, but accounted for 67 percent of offences. Indigenous offenders were much more likely to be chronic offenders than non-Indigenous offenders, with two-fifths (39.9%) of Indigenous offenders compared with less than one-fifth (15.8%) of non-Indigenous offenders classified as chronic offenders. Therefore, exploring whether some communities are more likely to generate chronic offenders than others may be an efficient way of targeting crime prevention interventions to reduce offending, crime, victimisation and Indigenous overrepresentation in the criminal justice system.

The geographic measure used to assess community location was the postcode where the offender resided when they first had contact with the criminal justice system. Each offender had their usual residential postcode recorded for each offence in the cautioning, conferencing, youth court and adult court datasets. These corresponded to POAs, which are the Australian Bureau of Statistics’ (ABS) equivalent of the Australia Post defined postal codes (ABS 2006a). The first recorded POA was selected in recognition of the importance placed on the early years of life from a developmental perspective and the cumulative nature of risk and protective factors.

While the ABS provides a number of standardised methods for measuring geographic location along with concordance files, POAs were used as the base measure of geographic location. POAs were used because the standardised geographical measures do not correspond directly to POAs. Alternating to these standardised measures would necessitate the random allocation of chronic offenders within single POAs to one of multiple standardised divisions. While probability derived concordance tables enable this to occur, doing so would introduce another layer of uncertainty into the data.

In Queensland, there were 432 POAs in 2006 (4000 to 4999). POAs differ substantially in both geographical size and population. The average size of a POA was 4,080.2 square kilometres (SD=16,621.7 km). The minimum area covered by a POA was just 0.4 square kilometres (4229—Bond University). However, the maximum area covered by a POA was 219,415 square kilometres. The POA that had the largest geographic size was 4871. This POA is located in far north Queensland and includes 58 different locations, one of which was the remote Aboriginal community Aurukun (see Appendix 1).

ABS statistics from the census were used to determine the population of each POA who were 16 years old in 2006 (ABS 2011a). These data were used because they were the most recent census data available at the POA level, covered the time when individuals born in 1990 would have been 16 years old and the average age of onset for offending was 16.21 years old. There was considerable variability in the base population of the 432 POAs, ranging from zero to 1,675 16 year olds (M=130.03, SD=187.14). POAs that had a population of 10 or fewer 16 year olds in 2006 (23.8% of postal areas) were excluded from analyses. This was because of the difficulties associated with small cell size and the random allocation process used by the ABS to prevent individual identification. After excluding these POAs, there were 329 POAs that had a population of more than 10 (M=169.42, SD=198.73, medium=96, maximum=1,675). The POA with the highest population of 16 year olds was 4350, which included the regional town of Toowoomba. It is obvious from these figures that the population is not evenly distributed across the postal areas.

ArcGIS was used to map the proportion of the population in POAs who were chronic offenders to determine whether some communities appeared to generate chronic offenders. POAs were categorised into four groups based on the proportion of the population who were chronic offenders using an average split (see Table 7). Additionally, the top 10 percent of POAs with the highest proportions of chronic offenders (33 POAs) were identified as locations where targeted interventions could be explored.

Because the project focused on the residential POA of chronic offenders when they first had contact with the criminal justice system, it was important to consider the potential role that offender residential mobility may have on limiting the usefulness of the findings for targeting interventions towards particular locations. The fifth research question was How residentially mobile are chronic offenders? To address this question, the number of times that chronic offenders changed POAs was examined.

Table 7 Proportion of population who were chronic offenders by number of postal areas
Category Population chronic offenders (%) Postal areas (n) Postal areas (%)
Nil 0 74 22.5
Low 0.1–4.72 150 45.6
High 4.73–9.09 72 21.9
Very high >9.09 33 10.0
Total   329 100.0

Phase five: Exploring which communities carry the burden of chronic offenders

The sixth research question was Which communities carry the burden of chronic offenders? As detailed in Cost of Offender Trajectory Groups in the Results section, individuals in the moderate and chronic trajectory groups cost, on average, between $58,116 and $262,057. While representing 3.8 percent of the population and 15.8 percent of offenders, they accounted for 68.6 percent of costs. Therefore, exploring whether communities could be identified that generate the most costly chronic offenders may provide additional information that will be useful for targeting crime prevention programs towards particular communities. This question was addressed by assigning the longitudinal individual costs to the POA where they first had contact with the criminal justice system. POAs were then ranked based on total cost and the top 10 percent most costly locations were examined. Total costs per chronic offender were established using the costing methodology described in Phase Three: Assessing the Costs of Offender Trajectories in the Methods section. These costs were aggregated for each POA. Once again, only the 329 POAs that had a population of more than 10 were included and costs were assigned to the offender’s usual residential POA when they first had contact with the criminal justice system. Across the 329 POAs with more than 10 individual, the average total cost of chronic offenders was $808,491 (SD=$1,441,216; range $0 to $14,041,855).