Where has all our employment gone?

After first release of Census, I wrote a blog about the higher quality of the Census data in 2011, compared to 2006. In general this is true, but unfortunately the second release has revealed an area where the quality of the 2011 Census is far far worse than 2006 or 2001. This is in the area of place of work, or work destination data, and it makes time series comparison very difficult for our users. This means we will be less able to rely on Census as an estimate of workforce numbers in 2011, and underscores the need for modelled estimates to give LGAs a better handle on their number of workers. Read on for more information.

The problem

The Census has always been an undercount of the true employment of an area. We discussed this in the blog Counting Employment – which figures to use? a while back. There are people who don’t state their workplace address and those who couldn’t be coded to an exact address at the local government level. Unfortunately this Census has raised the numbers of people in this category to a new level.

While there was a significant decline in those not stating their place of employment (due to the general decline in non-contacts for 2011), this has been more than offset by a five to ten-fold increase in those who have been coded to “Capital City undefined” or “State undefined”, meaning that they couldn’t be coded to an LGA of work.

The ABS claims that this is due to “greater precision” in their coders – ie. if an address was slightly ambiguous, rather than making a best guess at where it referred to, it was put in a statewide unknown category. And also due to the move to the new geography, which has SA2 units – much smaller areas to accurately code to. The problem is that as SA2s are roughly suburb/town level units, if the ABS couldn’t code accurately to the suburb or town, it would appear they didn’t even make an attempt to code to the (much more useful and important) LGA level, but just dumped it in the “State undefined” category.

The results

The results below show that the number of employed people who don’t have a Census place of work for has increased by a third, to more than 1.2 million. Reductions in “Not Stated” have been offset by huge increases in the undefined categories, while the numbers with “No fixed place of work have been stable.

The argument that it is due to coding to smaller geographic units in the SA2s is also probably not the whole storyl – if that was the case, Queensland wouldn’t have gone up by much, since a lot of the main employment areas in Queensland already had suburb-sized SLAs in 2006. But it’s increased as much as anywhere. Perhaps the quality of address coding is just a lot worse? That doesn’t bode well for the ability to get a Census form to everyone next Census which is moving to a mailout methodogy!

What it means

What this means is that in many cases, it is quite difficult to compare employment numbers between 2006 and 2011. A lot of LGAs have had employment numbers go down between 2006 and 2011, despite the number of employed residents increasing. It’s difficult to work out what is a real change in employment and what is simply due to the coding problems in 2011.

A great example is the Latrobe City Council in Victoria. In 2006, of the 28,318 employed residents, just over 2,000 had an unknown workplace address (either not stated, no usual address, or undefined Victoria). In 2011, the number unable to be coded is is 5,560, and State/Territory undefined is the second largest work destination after Latrobe City itself. This is close to 20% of the employed residents of Latrobe and represents an unacceptable level of error. According to Census, working population fell from about 27,000 to 25,000 in 5 years, while employed residents rose from 28,000 to 30,000. The difference is obviously the huge number of people in the state undefined category (most of whom would actually work within Latrobe City). There really should’ve been an attempt made to code these to LGA level.

It seems the problem isn’t so bad in metropolitan areas. The City of Monash, also in Victoria, but located in metropolitan Melbourne, has seen an increase in its employed residents with an undefined workplace from 8.3% to 11.9%, which is bad but not as severe as Latrobe. It also varies from state to state, and appears not so bad in WA – for instance the City of Belmont has about 1,800 people in going to an undefined workplace, about 11% of their employed residents, compared to 10% in 2006.

How to solve it

Fortunately for regional areas which are largely self-contained, the employment and industry numbers of residents are a perfectly adequate proxy for employment in the area. Our clients such as Bega Valley, Tamworth, Wagga Wagga, Swan Hill, Mount Gambier and Rockhampton have more than 90% of their workforce living and working in the same area, so they are far better off using their local resident employed numbers. Others such as Latrobe, Devonport, Albury-Wodonga and Toowoomba are largely self-contained but have some people coming in for work, so it’s not quite so good for them, but it may still be better to use the resident numbers.

These regional economies are the ones that live and die by their employment numbers. It’s far more important to a regional centre to know whether their employment is going up or down than a metropolitan area. Unfortunately the numbers are much worse in regional areas, but on the positive side, many of them are self-contained, so employed residents is a pretty good proxy.

For the metropolitan areas, fortunately the problem is not so bad, and it’s the movement between the LGAs which is more important. Just be careful when comparing levels of self-containment (proportion of local residents working locally) over time, as the increased number in the undefined categories may have an effect on this (since we assume that undefined means outside the area, for the purposes of that calculation). In profile.id we always show these numbers so you can decide whether it is affecting your results.

The other way to solve it is to not use Census data as a measure of employment at all. In economy.id, the NIEIR microsimulation modeling provides a good estimate of employment which factors back into the workforce all those with no usual address or location unknown, by using a combination of tax office and labour force survey data along with other sources to model the economy year on year. It has always been higher than the Census figures, and will be even more so once the new dataset comes out.

If you need help interpreting the Journey to Work figures for your area, please give us a call or send us an email.

Access more information about the Australian Census 2011.

Access the new profile.id sites and other population statistics for Australia, States, Capital Cities, Local Government Areas and suburbs at .id’s demographic resource centre.