|Page tools: Print Page Print All|
ADVICE ON REPORTING REGIONAL LABOUR FORCE DATA
The ABS recommends considering the following advice when interpreting and reporting regional labour force data:
To account for sampling variability, especially in regions with smaller populations, the ABS recommends that analysis of regional labour force estimates should be based on annual averages (as presented in Table 16(b) of Labour Force, Australia, Detailed (cat. no. 6291.0.55.001)).
The monthly Labour Force Survey provides timely information on the labour market activity of the usually resident civilian population of Australia aged 15 years and over. The statistics of most interest each month are the national and state and territory estimates of the number of employed and unemployed people, the unemployment rate and the labour force participation rate. The rate of change in the number of people employed is a key indicator of economic growth, and the unemployment rate is a key measure of unutilised labour. The participation rate reflects the percentage of the population in the labour force. The underemployment rate is an additional measure of increasing importance, of the extent of underutilisation of employed people.
The Labour Force Survey is designed primarily to provide accurate national estimates, with the secondary design objective of producing state and territory estimates. While the Labour Force Survey is not designed to produce regional estimates, these are compiled from smaller sample sizes at a lower level of statistical quality compared to those produced at state and territory and national levels.
Regional labour force data are published according to the Australian Statistical Geography Standard (ASGS) at the Greater Capital City Statistical Area (GCCSA) and the Statistical Area Level 4 (SA4) on a monthly basis in Labour Force, Australia, Detailed (cat. no. 6291.0.55.001). Each SA4 is designed to reflect, as best as possible, a discrete labour market within a state or territory, subject to the population limits imposed by the size of the Labour Force Survey sample.
It is also important to note that estimates are based on the place of usual residence, while respondents may be employed in a different region to where they live. This is particularly relevant for regions around capital cities, with workers often travelling across regional boundaries to central business districts, and labour market outcomes are more likely to reflect activity in these areas.
On a monthly basis, the Labour Force Survey samples approximately 26,000 dwellings which represents 0.32% of the Australian population. The sample is stratified across the regions of Australia to ensure a representative sample of survey participants and to minimise bias toward any one group of people. As a result, regions with lower populations tend to have fewer people sampled. Estimates produced from small samples are generally subject to proportionally higher sampling error, compared with estimates produced using larger samples. Data at SA4 level are also only presented in original terms, as it is difficult to estimate reliable seasonal factors at this level of detail.
Over time, large data fluctuations occur across most of the regional labour force data with relatively low populations. These fluctuations can be partly the result of local events (for example, the 2011 Queensland floods affected the collection of the labour force data for January 2011), but are generally due to sampling variability rather than changes in underlying market conditions.
SAMPLE SIZE AND RELATIVE STANDARD ERRORS
The Relative Standard Error (RSE) of an estimate is the inherent error of the sample as a fraction of the size of the estimate, and provides an indication of the percentage error likely to have occurred due the estimate being produced from a survey sample rather than the total population. The ABS publishes the RSE of each estimate produced from the Labour Force Survey to provide context to the estimates (see Labour Force Survey Standard Errors Data Cube (cat. no. 6298.0.55.001)). In published labour force data, any estimate with an RSE greater than 25% is marked with an asterisk to indicate that its value is subject to high sampling error and should be used with caution.
Graph 1 below provides a comparison between the unemployment rates for the time period January 2013 to May 2016 for Greater Melbourne and Shepparton. Graph 1 shows that the unemployment rate for Shepparton between January 2013 and May 2016 has almost all of its RSEs greater than 25%, while the RSE values for Greater Melbourne, which are based on a larger sample, are consistently lower at around 3%. Data for larger population areas, such as those separated into State, Greater Capital City or Rest of State and Territories, are likely to be affected by smaller sampling error, making point in time comparisons between these larger regions of higher quality.
GRAPH 1. RSE of Monthly Unemployment Rate, Greater Melbourne and Shepparton
INTERPRETING MOVEMENTS IN ORIGINAL DATA
All original labour force time series data (including labour market regions) consists of seasonal influences, irregular fluctuations and an underlying trend. An original data series with large irregular fluctuations can mask important underlying trends in the data. Data associated with regions of smaller population are more likely to exhibit large short-term fluctuations due to sampling error, and further statistical analysis of the data may be required before accurate conclusions can be formed.
Data at SA4 level are presented in original terms only, as seasonal factors are unstable at this level of detail. This can result in point in time inter-regional comparisons, using only the original data, being subject to influences from sampling error, seasonal influences and irregular components of the time series.
As an example, consider Graph 2 below which shows the unemployment rates of Greater Sydney and Illawarra over the period January 2013 to May 2016. Between December 2015 and January 2016, the unemployment rate for Greater Sydney rose from 4.6% to 5.3% and for Illawarra from 4.4% to 10.3%. This could possibly be a result of both regions experiencing higher unemployment rates, or an indication of an economic downturn. However, historical evidence shows that, in general, unemployment rates are seasonally lower in December than they are in January. Graph 2 shows that the Illawarra unemployment rate series was affected to a greater extent by irregular fluctuations than the same series for Greater Sydney. Patterns in historical data show that the unemployment rate for Illawarra fluctuates to a much larger extent in comparison to Greater Sydney, so this large increase in the Illawarra unemployment rate could be the result of an irregular, short term upward fluctuation.
GRAPH 2. Original Series, Unemployment Rates of Greater Sydney and Illawarra
SMOOTHING OUT SHORT TERM FLUCTUATIONS IN REGIONAL DATA
As described above, regional labour force data are more susceptible to irregular fluctuations in the original data and higher RSEs. However, the regional labour force data can be used to give an indication of longer term trends and analysis of regional LFS data should be undertaken on this basis. There are some simple methods that can be used to reduce the amount of variation, though these generally have some unavoidable disadvantages. The advantages and disadvantages of alternative methods are discussed in detail in A Guide to Interpreting Time Series - Monitoring Trends (cat. no. 1349.0).
A 12 month moving average is an intuitively simple method, which may lead to an improved interpretation (when compared with an unadjusted series) of the underlying trend movement as shown in Graph 3 and Graph 4 below. These show the difference in the unemployment rate time series from January 2013 to May 2016 for Greater Brisbane and Townsville, plotted using an unadjusted series and a 12 month moving average.
By applying an annual average to the original regional estimates, any seasonal influences are lessened and the monthly variation due to irregular fluctuations may also be reduced. However, the sampling error associated with regional estimates must still be considered before drawing any conclusions from the estimates, and the application of a 12 month moving average is unlikely to accurately or quickly detect turning points in the time series.
GRAPH 3. Unemployment Rates of Greater Brisbane; unadjusted and 12 month moving average
GRAPH 4. Unemployment Rates of Townsville; unadjusted and 12 month moving average
Starting with the July 2016 issue of Labour Force, Australia, Detailed (cat. no. 6291.0.55.001), the ABS will include a 12 month moving average spreadsheet (Table 16b). The original data for regional statistics will continue to be available in Table 16, to allow users to construct other moving averages, such as 3 month or 6 month averages, for regions with larger populations or for aggregations of multiple regions.
It is important to note that there are alternative and somewhat more complex methods for smoothing original regional series, such as comparing year-apart growth, and applying a 13-term symmetrical weighted moving average. However, a 12 month moving average is sufficient for most purposes.
In interpreting labour force regional time series data, it is important to consider both the strengths and the limitations of these types of data, including the relative standard error, before drawing conclusions based on the estimates. The regional estimates have, by design, unavoidably larger relative sampling error compared to the national and state and territory estimates, owing to their smaller sample sizes. Original data also contain seasonal influences and irregular fluctuations, which can mask the underlying trend of the data.
It is for these reasons that the ABS recommends that analysis of regional labour force estimates should be based on annual averages (as presented in Table 16(b) of Labour Force, Australia, Detailed (cat. no. 6291.0.55.001)).
The ABS wishes to acknowledge the assistance of the Queensland Treasury and their valuable contribution toward the content of this article.
FOOTNOTE: DEFINITION OF SAMPLING ERROR
Sampling error refers to the difference between an estimate for a population based on data from a sample and the 'true' value for that population, which would result if the whole population were enumerated. Sampling error is affected by a number of factors including sample size, sample design, the sampling fraction and the variability within the population.
FOOTNOTE: COMPARING REGIONAL DATA BEFORE AND AFTER 2013
Labour Force estimates have been published using ASGS regions since January 2014, and were backcast to October 1998. Estimates were backcast by determining from which SA4 each responding dwelling would have been sampled, had the ASGS been the geographical standard used for past Labour Force Survey sample designs. Backcasting labour force estimates by SA4s enabled a consistent time series of regional estimates to be published. However, because previous Labour Force Survey samples were designed using the previous geography standard rather than the ASGS, the creation of a consistent regional times series has had a slight impact on the quality of historical labour force estimates.
These documents will be presented in a new window.