20% discount is normally a good thing – but not with employment data

I recently met with a group of local government clients in NSW and we were discussing how to measure the success of their economic development activity by tracking change in the number of local jobs each year.

When I made the statement that the census undercounts local (LGA) employment by up to 20 percent, they were shocked. “Was I sure about this?” They asked. “Does everyone know about this? What are we doing to make sure people in economic development know?”

Let me be clear, the census is an incredible source of knowledge about local communities. In most cases it is a very robust and reliable source of information. And five yearly updates are frequent enough for most analysis. But it does have its limitations.

Census limitations on counting the workforce

The census is best at measuring the number and characteristics of people who LIVE in a particular area, because the census form is completed at an identified residential address. But how well does it capture the number of people who WORK in a local area? It is very good at doing this at a national or state level because almost everyone works in the same state they live in. But many, even most, people don’t work in the same LGA as they reside, so the census asks them to state their workplace address. This is a free-form field. Some people complete it, some don’t, and others provide incomplete or illegible information that can’t be accurately coded to their place of work, while others provide addresses that relate to head offices, rather than where they actually work. Still more genuinely have “no usual address” – that is, they work in a variety of locations depending on the day of the week or time of year, and though they may mainly work in the same area, they can’t give an address and therefore can’t be coded.

In 2011, the ABS also had a lot of issues with correctly allocating addresses to work locations, mainly due to the shift to the new geographic areas which de-emphasise local government.

The number of responses accurately coded to a work place address in 2011 is about 88% of those who state that they work. Combined with another 5.6% of population who don’t even answer the question on employment, that means on average, local employment is undercounted by about 18%.

Large Census undercounts for some industries

When you start breaking this down by different industry sectors, the story gets more complex. For some industries the undercount is worse. People who work in the construction industry, for example, work in many different locations and may not know which location to state. Many will state a head-office location rather than the actual place the construction is occurring. If an LGA has a lot of construction work taking place the employment undercount can be significantly higher than 20 percent. When the undercount in one industry is more marked than others it can skew the relative importance of different industries in an area.

This table shows the percentage of people in each industry who have an unknown work location for one of the reasons above. Note that this isn’t the total undercount because it doesn’t include the 5.6% of population who don’t answer the employment question. Also note that many of these people ARE coded to a STATE of work, just not to an LGA.

In the extreme case, for almost one-third of Construction workers across Australia, we don’t have a place of work! So there is a need to make some adjustments to get a better estimate of the actual number of workers in any industry.

City of Wanneroo – an example

The Census records 31,823 workers with a workplace address which could be coded to the City of Wanneroo, on Perth’s northern outskirts. This is one of the fastest growing areas in the nation, and it is not surprising that Construction (with 4,807 workers) is the largest employer, though only just ahead of Manufacturing.

The modelled data, which adjusts for the undercount in different industry sectors, shows 44,002 workers in the City of Wanneroo for 2011/12. Construction remains the largest employer, but by a bigger margin, now having 8,631 workers, or 19.6% of total.

This shows the importance of using alternate sources to adjust the Census data.

Drawbacks of using only Census data in your economic analysis

Many economic models use Census counts of employment as the base data from which all other data sets are modelled. This means that they:

1. Undercount employment by up to 20 percent on average (more in some areas);

2. Can misrepresent the underlying structure of the economy because some industries have a higher undercount than others;

3. Only update the base data every five years when there is a new census, which means any structural changes in the economy go unmeasured, as do significant job gains or losses. For example, the recent announcement that Ford will be closing it’s Australian manufacturing plant in Geelong with a loss of 1,200 jobs will not be reflected in Geelong’s count of local jobs until the next Census.

4. Any impact assessment which looks at the flow on effects of new jobs across the rest of the economy is using multipliers based on an initial undercount thus magnifying the problem.

In addition, some economic models also exclude people whose workplace address is known but whose industry is not stated. This adds about a further 1% undercount.

What is the solution?

The best source of data about local employment by industry is still the census. The number of jobs Australia wide is known. Over eighty percent of them are accurately coded to an LGA. But the missing 20 percent needs to be allocated to the right LGAs and the right industries within those LGAs. This requires specialist modelling techniques based on alternative data sources and a passion for understanding local economies.

.id are experts in Census data, so we understand its shortcomings. We also know that creating local employment is one of the most important objectives of economic development, so measuring it accurately and frequently is paramount. Our challenge was to find an economic model that we could confidently offer to our local government clients that resolved the undercount problem.

We selected NIEIR (National Economics), well known to local government from The State of the Regions report they produce each year for ALGA. Unlike most economic models which apply input-output modelling to the census count of employment, NIEIR undertake micro simulation modelling with a range of data inputs to allocate the missing 20 percent of jobs to an LGA before undertaking input-output modelling. These data sources include the monthly Labour Force Survey by ABS (which doesn’t suffer from the same level of undercount, as questions are asked of respondents by trained interviewers), Centrelink estimates of employed and unemployed population, and Tax office counts of employment by industry which are based on actual tax returns.

A micro simulation model uses additional sources of data to make decisions about where to allocate each industry’s missing workers.

Once the model is in place they can also adjust it based on local factors such as the Ford factory closure each year.

It’s not perfect. It is still only a model of reality. But it is much better than just accepting the undercount. And it provides local government with an opportunity to add their local knowledge to the model.

From our discussions with economic development officers in local government, we have come to realise that this is a little known issue that can have a significant impact on decision making. If you found this helpful, please share it with your colleagues.

Next we’ll be looking at how the number of full-time vs part-time jobs impact on economic modelling.