1351.1.55.160 - Exploration of state space modelling approaches for statistical impact measurement in ABS time series: The Labour Force Survey as a case study (Methodology Advisory Committee), June 2018  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 25/09/2018  First Issue
   Page tools: Print Print Page Print all pages in this productPrint All

Executive Summary

The ABS is embarking on a transformation program, which includes a change in ABS systems (editing, imputation, estimation, etc.), a use of different collection modes and a change in sampling frames etc. Once this transformation is completed, it is expected to deliver positive changes to the production of official statistics. However, there is also a risk of introducing impacts on some ABS time series. The ABS is working on developing methods to measure, and where needed to adjust for any impacts on such time series.

This paper discusses the usage of state space modelling approach to measure statistical impacts under two scenarios: 1) having a small parallel collection and then incrementing (phasing-in) the new approach to the survey, 2) no parallel collection is available and a new approach is introduced gradually. We consider a few models in this study, including the difference model for parallel collection, phase-in model with and without Kalman filter initialisation using estimated impacts from parallel collection and finally, the full model that utilises both parallel collection and phase-in period information. The models’ performance is evaluated from the results of a simulation study for the Australian Labour Force Survey (LFS). Full model with parallel collection performs the best in terms of the power of detecting overall impacts when compared to the other models.