The modelling method has explicitly considered the effects of seasonality and survey sampling.
Residual seasonality was included in the unemployment model because of the different seasonal patterns that occur in unemployment relative to DSS JobSeeker recipients, particularly around the summer holidays when labour market activity is reduced. For employment, no residual seasonality was detected or required.
The models also include a sampling error term, which is modelled such that the temporal correlations (and variances) induced by the rotating panel design of the LFS are captured. By modelling the direct survey estimates and their associated sampling error covariance structure the complex sample design of the LFS is properly accounted for. The models also include random effect terms for both areas and time points. These terms help to account for differences between areas and time points that the auxiliary data and seasonal terms could not capture.
The models borrow strength across time by the inclusion of the temporal random effects and by the modelling of the correlated sampling errors. These features improve the accuracy of the modelled estimates, particularly increasing the stability of movement estimates.
The models assume a roughly constant relationship between LFS variables and the auxiliary data over time. When there are government policy changes and/or unusual events such as COVID-19, these relationships may change significantly. Time series corrections have been applied to account for this, particularly COVID-19 effects on unemployment.
The models have been applied separately for each of the larger states. For smaller states, SA4s from other states have been combined into the models, as deemed conceptually and empirically appropriate. In particular:
- Tasmania also uses Victoria in modelling
- Northern Territory also uses a selection of outback SA4s in South Australia, Western Australia and Queensland
The Australian Capital Territory (ACT) is only comprised of one SA4. Given the estimates from these SA4 models are designed to be additive to the existing national and state and territory level estimates, the ACT data in the modelled SA4 series will continue to be the same as the territory level direct survey estimate.