This edition of the CVS publication presents, for the first time, pooled data on victimisation and reporting estimates at the state and territory level. Data pooling is a statistical technique in which data from multiple survey cycles are combined together to produce more reliable estimates. This effectively increases the sample size of the data set and allows for the production of more robust estimates by decreasing the associated sampling error. For the CVS, this has particular utility for improving the reliability of estimates for less populous states and territories, and lower prevalence crime types such as robbery and sexual assault. The pooled data have been provided in addition to the existing time series estimates and can be found in Tables 28-33.
The pooled estimates in this release were derived by combining data from consecutive CVS reference periods (survey cycles) and dividing the weights by the number of cycles that were pooled. The estimates therefore do not correspond to one particular survey cycle but instead are an average of the estimates over the pooled reference period. Two-year pooled estimates were deemed suitable for the majority of personal and household crimes. In the case of sexual assault however, three survey iterations were required to produce meaningful estimates at the jurisdiction level, due to the relatively low prevalence of this crime in Australia.
CVS data have been pooled based on an assessment of the survey’s comparability and consistency over cycles. Each CVS iteration is conducted on an independent sample of the same population, and the population characteristics and variables of interest have not changed substantially from one survey to the next. Survey questions regarding victimisation and reporting have remained consistent since the first CVS in 2008-09. Successive CVS iterations have also had similar survey design, scope and coverage, enumeration period and weighting method, indicating a high level of conceptual similarity between surveys.
The pooled reference periods have been labelled according to the survey cycles they are composed of. As such, the 2008-10 pooled period combines data from the 2008-09 and 2009-10 CVS cycles and therefore spans the time interval from July 2008 to June 2010. However, while the pooled estimates draw on data from multiple years, they do not reflect the incidence of violence over the pooled period but instead are based on respondents’ experience of crime over the 12 months prior to interview. For example, the 2018-20 pooled victimisation estimate for physical assault averages the number of persons from the 2018-19 and 2019-20 CVS cycles who reported experiencing physical assault over the last 12 months.
To provide data users with the greatest amount of flexibility, rolling time series estimates have been provided in the pooled data tables. The rolling estimates combine consecutive periods of a survey so that a survey year can be used in multiple estimates. For example, the 2009-10 CVS cycle has been combined with the 2008-09 cycle to produce the 2008-10 pooled estimate, and with the 2010-11 cycle to produce the 2009-11 estimate. While rolling estimates produce the greatest amount of time points (compared to a non-overlapping approach, in which a survey’s data is used for only one pooled estimate), caution should be used when comparing estimates with overlapping data. These estimates will not be independent due to the overlap in survey years, which can make analysis and interpretation of the data more difficult. The state and territory section of the commentary includes comparisons between the first (2008-10) and last (2018-20) pooled estimates in the time series. Comparisons to the most recent non-overlapping time period (2016-18) are also discussed.
The estimates in the pooled data tables have not been perturbed, in contrast to estimates in the other publication tables. Instead a different confidentiality measure has been applied and some cells have been suppressed (though included in the totals where applicable), to minimise the risk of identifying individuals. The use of a different confidentiality measure means that some pooled estimates may be higher or lower than both of the previously published single-year estimates that make up the pooled data.