Prior to releasing the September Labour Force Survey (LFS) results this month, the Australian Bureau of Statistics (ABS) announced they were taking the unusual step of dropping the seasonally adjusted Labour Force estimates for July, August and September. Many users of the monthly survey results have since expressed concern about the quality and reliability of the estimates. Some of the commentary has suggested the recent issue with the estimates was due in part to budget cutbacks to the statistics agency. However, the recent volatility in the seasonally adjusted estimates is more likely to be associated with the changes the ABS has made to the LFS program rather than any budget pressures they are currently facing.
What’s changed and why?
To understand how the survey program changes impact on the seasonally adjusted estimates it is important to first understand what changes have been made to the survey program and why they were made. In July 2012 the ABS announced the outcomes of their Labour Household Content Review (Cat. no. 6107.0). The review was undertaken to ensure the ABS labour statistics continue to be relevant and to ensure the statistics are collected in an efficient manner. The changes most relevant to this discussion are the changes to the number and the timing of the labour supplementary surveys.
Supplementary surveys are run at various points through the year. Questions from these surveys are asked after the completion of the monthly LFS. Different groups of people are asked questions specific to their circumstance. For example, in August, prior to the changes, people who were employees were asked a range of questions about the jobs they had. This included questions about their earnings, whether or not they were members of a trade union and questions about benefits they were entitled to.
From July this year, the ABS has reduced the number of labour supplementary surveys they run each year from five to two (see Table 1 and Figure 1 for more details). Content from the previous surveys was combined into surveys run in either February or August. The changes are designed to maximise the value of the information collected and ensure the program is more coherent (for further information see Information paper: Outcomes of the Labour Household Surveys Content Review, 2012).
Why did the ABS decide to drop seasonally adjusted estimates?
The ABS produces three different types of estimates from the Labour Force Survey, they are:
- original or unadjusted estimates
- seasonally adjusted estimates and
- trend estimates.
The original estimates are unaffected by the changes the ABS has implemented. The seasonally adjusted estimates and trend estimates are affected.
The issue the ABS faced when making these changes is that the supplementary surveys contribute to seasonal effects on key estimates such as the unemployment rate and estimates of employment in the LFS. A seasonal effect is a statistical pattern that can be predicted and the size of its effect on the data can be estimated. The seasonal adjustment process the ABS uses removes the seasonal effects from the original data so the underlying trends can be identified more easily. By moving the labour supplementary surveys around and changing the content of the surveys the normal seasonal effects for the affected months have changed. For example, in August the estimate for employment usually decreases. In August this year, however, the estimate for employment increased by 32,100 (Figure 2). The only other time the employment estimate increased in August was when the normal August supplementary survey was shifted to July in 1991.
Attempting to anticipate how seasonal patterns have altered in response to changes to the survey program is extremely difficult. This task is complicated as the seasonal pattern for any month is influenced by a range of factors, not just the timing or content of a supplementary survey.
The difficulty of predicting changes to the seasonal effects is demonstrated by the surprising results in August. While the core components of the August supplementary survey remained in place in August this year (Figure 1), the seasonal pattern of a decrease in employment did not (Figure 2). It was only after having three months data that the ABS could be confident the changes put in place had likely changed the seasonal pattern and thus using the normal seasonal adjustment process would be inappropriate.
What is the ABS doing?
The ABS has advised users they are undertaking a review into the series to better understand why the series appear to have shifted away from their usual seasonal patterns. However, it is unclear at this stage how other months likely to be affected by the changes will be treated. The months most likely to be affected by the change in the survey program are February, July, August, September and November.