Many New Zealanders are interested in how crime rates in their neighbourhood compare with rates in other parts of the country. At present, the main public source of information on geographical variation in crime is offence statistics for police areas and districts, available on the Statistics New Zealand and New Zealand Police websites. These are the numbers most commonly reported by the media, including the Press article cited above. Police areas and districts are suitable for some sorts of comparisons, such as examining police workloads or performance. But to answer questions about which areas have the most crime, the geographical units that are used should align with standard definitions for these areas.
In New Zealand, the police station is the smallest geographical unit we have comprehensive time series of offence statistics for. Detailed, consistent electronic data at the police station level exist back to 1994, permitting trends to be analysed over almost two decades. By grouping police stations into larger units that approximate standard geographies such as territorial authorities and regional councils, it is possible to estimate offence rates for these geographies. One study that takes this approach is the Quality of life in twelve of New Zealand’s cities 2007 report (Quality of Life Project 2007: 91–98), which presents offence rates for burglaries, violence, sexual offences, car offences, and drug and antisocial offences at the territorial authority level. Another is the ‘Regional indicators’ section of the Ministry of Social Development’s Social report, which presents rates for all offences at the regional council level.
Aggregating police station data is a promising approach to creating time series for regional crime statistics. However, any attempt to produce general-purpose statistics from administrative data inevitably requires trade-offs between accuracy, timeliness, and transparency (Rees 1986; Freedman et al 2008; Gregory et al 2010). It is important to assess these tradeoffs, to help inform methodological choices, and to provide users with information about data quality.
In this paper we investigate three simple rules for constructing regional crime statistics out of police station data. The rules assign police stations to target geographies based on population, land area, and a combination of the two. The target geographies that we use are regional councils, territorial authorities, and main urban areas. We evaluate our new geographies by calculating the extent to which population and land area are misclassified, and by examining whether different rules for allocating police stations have a material effect on regional crime comparisons.
We find that the allocation rule that uses both population and land area performs best. The difference in performance is generally small, however, with all three rules performing well, except for smaller urban areas. We suggest that it would be feasible to construct crime statistics for regional councils, territorial authorities, and large urban areas, but not for small urban areas.
Purpose
The only publicly-available information on the geographical distribution of crime in New Zealand is offence statistics for police administrative units. We investigate whether existing data can be used to construct geographical crime statistics that correspond to regional councils, territorial authorities, and urban areas.
Methodology
We build experimental output geographies from police stations, the smallest administrative unit for which there are long time series of offence statistics. We develop three rules for assigning police stations to the new geographies: one based on population, one based on land area, and one based on both. We assess the performance of these rules by calculating the proportion of national land area and population that is misclassified, and the number of target units that do not receive at least one police station. We also look at whether regional statistics on serious assaults are sensitive to the choice of allocation rule.
Key Results
The new output geographies approximate the target geographies well. For instance, our preferred rule assigns 96 percent of the national population to the correct territorial authority. Moreover, a case study of serious assaults suggests that most regional crime statistics are not sensitive to the choice of rule.