The previously posted Cairns Chamber presentation from Bill Cummings includes some strong opinions on survey sample data:
Unfortunately, there is another series based on sample surveys that causes confusion – the ABS Labour Force series is published each month. Just how bad the series is, is illustrated by comparison with Census data when it came through.
Source: Cummings Economics
Did have a quick look at this previously and put together a few numbers which I haven’t gone back to check which may have been wise but anyway:
The numbers for the ABS Labour Force Survey (LFS) numbers are taken from August each year which is the month of the census. The ATO data (also from ABS) is derived from tax data for the prior financial year.
The Cummings difference for Cairns appears to be derived from the ABS recommended 12 month average for the (LFS) data. Had he used the Conus Trend the decline would have almost halved to 5,100. The original monthly data for August would have been lower too but that’s as much a fluke as anything given the notorious volatility of the regional LFS data.
The ATO data here is derived from two different series which represent categories of employment . I don’t claim to be on top of the nuances of these different series. It isn’t the number or the discontinuity that matters but the trend. I can’t think of any reason why the trend in this data should not be consistent with employment in Cairns over the period. Previous posts on this data: Personal income down to SA2; ATO income data and Cairns City.
Similarly we shouldn’t focus on the difference in number of employed between the Census and LFS: Differences between the LFS and the Census, which includes comparisons for the 2011 Census. These will be different and a quick look suggests an expectation of a similarly higher number at both Australia and Queensland level for the LFS at the 2016 Census. It’s the 2011 – 2016 period where the difference appears anomalous for Cairns SA4 between different data series.
In timely fashion this month Ricardian Ambivalence rose from a long slumber to remind us that “the survey isn’t designed to measure the number of people with jobs. It is designed to measure the unemployment rate.”
Explained in more detail back in 2012: It’s the unemployment rate …
The household survey is NOT designed to measure the number of jobs. It is designed to measure a bunch of key ratios, including the unemployment rate. Multiplication of population forecasts by the ratios yields the so-called jobs number. Population over- and under-estimates are corrected for by smoothing subsequent population estimates toward the revised estimate, and this may distort the derived levels.
The ABS Labour Force Survey problem is not that any number could be wrong but that at regional level employment data may be misleading not just over short periods related to sample volatility but over extended periods.