Skip to main content

Exploring the oddities and peculiarities of Census 2021 Change layers

Overall Population Increase but many areas have decrease

As generally expected, there has been an overall population increase of 8.6% across Australia between the 2016 and 2021 censuses. One might expect to see the population to increase steadily across all areas but that does not seem to be the case.  

Exploring the latest OneMap Change layer for Population, which shows census-on-census population changes at the meshblock level, we can see that many areas of the country have seen significant declines in population. 

Shown below for example in the Sydney region, we can see wide swaths of population decline (in red) in areas like Paramatta, CBD, and Kingsford. Focusing in on Kingsford more closely and analysing the data we see that in area there has been a ~14% decline in population, which equates to a loss of roughly ~5,500 people. 

Census Sydney Population Change
Parramatta Population Decline
Kingsford Population Decline
SydneyPopChange Kingscross
-14.2% Decline in Population


As we explored in a previous post about Census 2021 data, there has been a general trend in Median Household Incomes increasing, with some areas seeing massive Median Household Income increase (Highett, VIC +39%). One might be inclined to associated this with a change is housing stock, which would bring with it a new demographic.  

Some apparently developmentally ‘stable’ areas however are showing over 100% increases in Household incomes with no obvious developments or changes in the housing stock. 

svg+xml;charset=utf — OneMap

In this area of Oakleigh, VIC (SA1: 21205132601) the Median Household Income has risen by 101.5%.
A doubling of the Median Household Income
since the previous census in 2016, with no obvious change in the area between the two time periods.
This could, of course, be explained by a change in demographics in the area, a change in jobs for the residents, or indeed the influx of one very very High Income earner into the area… one for the demographers to ponder. 

svg+xml;charset=utf — OneMap
+101.5% Median Weekly Household Income Increase

Rents Down in Places

Below we see the pattern of change in Median Weekly Rent values between the 2016 and 2021 Censuses in Brisbane. What is immediately apparent are the areas of decline in Median Weekly Rents in Central Brisbane, shown in red below. 

Looking at the numbers we see that in a 1km area within the Brisbane CBD the Median Weekly Rent has declined by 5.3% from 2016 to 2021. This is in spite of an overall Increase of 6.6% in Median Household Income in the same area. 

This, it must be pointed out, is the exception and not the rule, and we can see in most areas Median Weekly Rent has Increased, and in some cases quite significantly! 

svg+xml;charset=utf — OneMap
-5.3% Change in Median Weekly Rent

Reasons For Change

This particular area of St Kilda, VIC, saw an increase of +463.64% in population but sits alone in a sea of population decline, what’s going on here? 

svg+xml;charset=utf — OneMap

We can use a way back machine, in the form of the NearMap Timeline or MetroMap Timeline, within OneMap to view Aerial Imagery from a point in time.  
Logically enough, with those tools at hand we can see that at the time of the last census in 2016, the meshblock in question was a building site. The Site of a building, that when completed would house up to 51 people, and account for the 463+% increase in population in that same location. 

svg+xml;charset=utf — OneMap

Notes on Data Processing

In order to generate these difference layers we gathered the processed 2016 Census data along with the newly processed 2021 Census Data. Step one of all this work is of course to assign each record to its corresponding spatial location, we used the smallest/most detailed data available. Population data was assigned to meshblocks and Median Household Income was assigned to SA1 for example. 

Comparing data census-on-census is only possible where there is a common unchanged boundary across both comparison years; it is only possible to compare like with like.
In areas where the meshblock or SA1 boundaries were new, retired, or altered, it was not possible to compare change. In these cases, no comparison is given, you might notice some gaps in the map, that is what is at play here.
The OneMap analyser tool results showing % change only factors in meshblock or SA1 boundaries where a comparison is indeed possible, ensuring that the results are the shown for where there exists raw data for both input census years. 

We decided to use bubbles as the mapping style as it allows us to show both the direction of change (+/-) as well as scale of change at a glance. 
This idea was inspired by some previous seen work such as this from Jonas Schöley:

svg+xml;charset=utf — OneMap

And the Maps featured here in the Washington Post:

svg+xml;charset=utf — OneMap
AUTHOR: Michael Cushen