Eric on geography and Christian life

Counting Christians in the Australian census

This article is about estimating how the proportion of active Christians varies across Australia, using data from the 2011 census, along with other available data such as from the National Church Life Survey. A number of things to note:

  • My intended readers are thinkers and leaders in the Christian community.
  • I’m not that careful with my terms, but I talk about “Christians” or “active Christians” meaning people who follow Jesus Christ. Census religion statistics are something much different, but I endeavour to use the latter to estimate the former.
  • I am more interested in comparing different areas that counting absolute numbers.
  • There are lots of technical details. You don’t need to understand them all, but I include them for completeness.
  • This may change or expand in the future if I do further analysis.

Starting point

Much of the books of Numbers and Joshua are concerned with counting the people of God and mapping out the land. William Carey’s 1792 Inquiry listed the known countries with population and religion data. Today lots of work goes into researching and documenting the proportion of Christians living in each country and people group. There are several such projects on the world scale. The questions are: “How is the mission going? Where are we up to? Where is there more work to do?”.

For several years I have been pursuing this topic as it applies to Australia, in particular Adelaide. Which parts of the city have more active Christians? Does this reflect settlement patterns of Christians? Does it indicate where the gospel is or isn’t spreading? This is my starting point.

Why the census?

So we to know what proportion of a population are following Christ. An obvious measurable indicator would be church attendance. There are some attendance stats available, which I’ll come to later, but they are not always comprehensive.

There is great value in census stats (I’ll get to the drawbacks shortly). The ABS releases gigabytes of raw figures on many topics, categorised in different ways and counted up for every part of Australia. So even if a demographic question you are asking is not directly addressed by the census, at least the answers you get will be comprehensive, covering the whole country in a lot of geographical detail. When you have a complete set of figures, it is easy to run calculations and generate maps. And because the census covers the whole of Australia, my scope has expanded from my home city of Adelaide to the whole country.

As I am a software developer, I have always made software tools to analyse the statistics. Since 2008 I have used Mappage, my multi-purpose web application, to show tables listing figures for different areas and colouring in a Google Map accordingly. Lately I have added a scatter plot, which I’ve found is my favourite way to look at the stats. Although a scatter plot is designed to compare two variables (which I’ll do later), I mostly use only one, with all the bubbles spacing themselves evenly in a rough geographical order. If you run Mappage with this link (no longer works), you will have a row of links that call for different scatter plots. When this page has a number in square brackets, you can click the link with that number to see the graph that shows the data I am describing.  (You’re able to play with other settings, as well.)

As the census data is organised geographically, so are my calculations. But missiologically, geography is just one dimension when we look at the spectrum of mini-people-groups that make up society. Age, gender, ethnicity, occupational, social, and other dimensions are often more relevant, and we know that Christians are plentiful in some and rare in others. Often a small geographical imbalance may be a clue towards a bigger imbalance along some other axis.

First attempts at using the data

The first thing we see [1] is that 61% of Australians identified as some sort of Christian, with 7% for other religions. We know that this 61% is many times more than the active Christians (weekly church attendance is around 8%). Religion in the census is more about which cultural group one belongs to and where one goes to church on special occasions. So when we see that the Total Christian figure ranges from 43% in Sydney – City & Inner South to over 75% in Sutherland and parts of rural NSW, how do I estimate the relative strength of the Church in those areas?

The NCLS Occasional Paper 2001 Church Attendance Estimates lists several denominations with weekly attendance estimates, census figures and the ratio. This attendance ratio, which I will use extensively, is over 70% for Churches of Christ and Pentecostal churches, so we can assume most people indicating one of these in the census are involved with a church. At the other end of the scale, attendance in Anglican churches in a typical week is less than 5% of census Anglicans.

I have found it useful to break the 30 or so census religion categories into five groups (some small categories don’t fit into these groups – I don’t discuss them here). None comprises No religion and Not stated. Other Rel aggregates all the non-Christian religions. The Christian denominations are broken up into Cath Orth, Trad Prot and Strong Prot as follows:

Cath Orth – 28% Trad Prot – 26% Strong Prot – 6.1%
Catholic (25%), Eastern Orthodox (2.6%), other small non-protestant categories Anglican (17%), Uniting (5%), Presbyterian & Reformed (2.8%), Lutheran (1.2%) Christian nfd (2.2%), Baptist(1.6%), Pentecostal(1.1%), Churches of Christ, Salvation Army, Seventh-Day Adventist, Brethren, Other Protestant
These denominations have attendance ratios of less than 20%.
They might broadly be described as traditional and are often linked to particular ethnic groups.
They generally practice infant baptism.
Many people will identify with the church on the basis of their cultural background or their baptism, despite not regularly attending church.
They had a box to mark in the census form.
These denominations have attendance ratios of greater than 30%.
They might broadly be described as evangelical.
They mostly practice believers’ baptism.
They did not have a box to mark in the census form (except Baptist), respondents had to write it in. Christian nfd counts those who simply wrote “Christian” – after all, the census asked “What is the person’s religion?”.

So when the statistics show None or Other Rel, we know we generally don’t have Christians. When they show Strong Prot, we do have Christians. When we have Cath Orth and Trad Prot, that’s where we have the most uncertainty. With weekly attendance rates of 5-15%, translating to ongoing church involvement rates of maybe 10-30%. So when an area shows a large figure for Anglican or Catholic, we don’t know whether they include many active Christians or not. Once again, further data like church attendance figures would be most helpful. In particular, there is a very strong correlation between Trad Prot and the proportion of people born in Australia.

Having defined those five groups, a quick look at what proportions we find across Australia and what have been the changes from 2006 to 2011.

Group 2011 06-11 Places with the highest proportion Places with the lowest proportion Movements 2006-2011
None [2a] 27% +2.1% Inner Melbourne 38%, S Adelaide 38% SW Sydney 8.7%; Blacktown 12.5% +5% in Tas
Other Rel [2b] 7.6% +1.6% Parramatta 30%; SW Sydney: 27%; Inner SW Sydney 23% Much of rural Australia 1-2%
Cath Orth [2c] 30% -1.2% NW Melbourne 46%; SW Sydney 46% Rural SA & Tas 16-18%
Trad Prot [2d] 27% -3.1% New England & NW NSW 48%; A few rural areas over 40% SW Sydney 11.2%; Much of Sydney and Melbourne 13-17% -5% in Tas
Strong Prot [2e] 6.5% +0.5% Toowoomba 11.4%; Outback NT 11.1% Eastern Suburbs Sydney 2.4%; Inner Melbourne 2.7%; Inner S Melbourne 2.9% Increases everywhere

The Strong Prot group is of particular interest, because so far it is our only strong indicator of numbers of Christians. The proportion varies quite a bit around Australia. The inner parts of Melbourne and Sydney are lowest. Toowoomba is highest, along with outer parts of Brisbane, and most of Queensland scores highly. Outer Eastern Melbourne is 9.4% while Victoria only averages 5.2%. The north of Tasmania scores much higher than the south.

So we have a fair idea about the distribution of about a quarter of Australia’s Christians. Before we look for the others, I’ll digress and explain some things about Australian statistical geography.

Census geography

This is the hierarchy regions the ABS use:

Code  Type of region How many in Aust Typical population in each Example (where I live)
AUS whole of Australia 1 22 million Australia
STE  state or territory 9 millions South Australia
SA4 Statistical Area 4 88 hundreds of thousands Adelaide – West
SA3 Statistical Area 3 333 tens of thousands Charles Sturt
SA2 Statistical Area 2 2,214 thousands Woodville – Cheltenham
SA1 Statistical Area 1 54,805 hundreds (part of Cheltenham)

So Australia is divided into states and territories, which are divided into SA4s, which are divided into SA3s, etc. Everything measured in the census is available for SA2s and above. Basic data is available for SA1s. As well as this hierearchy, ABS releases basic data for other regions – local government areas, suburbs, postcode regions, electoral districts – but these are just aggregated from the SA1s.

The scatter plots work well with 50-100 bubbles, so they typically show all the regions two levels down, ie all the SA4s in Australia, all the SA3s in a state or all the SA2s in a SA4. The figures I’ve quoted so far are for SA4s. The further down you look, the more extremes you may find.

The sum-product method

Having mapped out the distribution of about a quarter of the Church in Australia, we will try and get some idea about the rest. One of my favourite measures uses what could be called the sum-product method. For any area, we take the census figures in their denominations, multiply each by the attendance ratio for that denomination and add them all up. This gives a rough measure for the Christian-ness of an area. It could be the estimate of weekly church attendance for that area (actually an overestimate, as I’ll explain later). The actual number of people connected with the churches might be nearly twice this.

In the case of the Anglican Church, we have diocese-level attendance ratios to multiply by. Newcastle, Brisbane and Perth are all below 3.5% while Sydney and Armidale are over 6%.

This sum-product score [3] is the measure I’ll use for the rest of this article. South East Tasmania and the inner areas of Sydney and Melbourne are lowest, with scores below 7%. Toowoomba is by far the highest with 12.4%, a whole percent higher than the next two SA4s, Blacktown and Baulkham Hills & Hawkesbury.

The biggest increases for this measure since 2006 are outback NT (1.2%), Mandurah (0.8%) and the WA Wheat Belt (0.8%). WA as a whole increased by about 0.4%, much of which is due to immigration from South Africa. The biggest decreases are inner Melbourne and a number of inner parts of Sydney (0.5-0.6%).

I think this method gets us some way to approximating the relative strength of the Church, but there is plenty of uncertainty. The total using this method is affected most by the Strong Prot figures, because of their high multipliers. It would be correct if the ratio of active to nominal Christians in each denomination was uniform across the country (or diocese, where we have ratios for them). But in an area where many of the active Christians are Anglican (north shore Sydney, perhaps), the score will be less. A shift of members from Uniting to Pentecostal or Christian nfd would result in a higher score, with the same number of Christians.

If, instead of the active to nominal ratio keeping uniform, the Anglo-Australians of no faith divided in a roughly uniform way between No Religion and Anglican (and other categories), and the way they divided was roughly uniform – then the overall scores would diverge more.

Another source of error is that my multipliers are from 2001. If I can get them for 2011 I’ll use those instead.

I’ll use this sum-product score later and compare it with other demographic factors.

Complications – Christian nfd

Much of Christian nfd would be pentecostal church attenders, and much of the rest might be of independent churches, house churches and other Strong Prot churches that didn’t have a box on the census form. For this reason, the multiplier used for Pentecostal and the other Strong Prot figures are reduced from their higher levels (eg from over 70% to 60%), but Christian nfd is given a multiplier of 60%.

Complications – Not Stated

In every topic of census data, there is a Not Stated category, where no answer was given. Typically this means the census form was not filled out at all for that person, or was returned not completed. This is usually over 5%. When I am calculating percentages, I usually subtract the Not Stated figure from the denominator.

The census form specifically says that the religion question is optional. Consequently there is a larger Not Stated figure. In the 2006 data, there was a strong correlation between No Religion and Not Stated, so I thought it was fair to consider Not Stated as another way of saying No Religion, ie a multiplier of zero with no adjustments. This wasn’t that accurate. There was a larger correlation between Religion not stated and Not stated in other topics, indicating the non-response rate. Because of this, places like Robe ranked as less Christian, while inner city areas already higher in No Religion and with a big non-response rate became the outliers.

In 2011, the Not Stated proportion dropped (11.2% to 8.6%) and No Religion rose sharply. There was no longer much correlation between the two, and a stronger correlation between Religion Not Stated and Not Stated for other topics.

So it is more important to deal with this figure fairly. My current way of handling it is to drop the denominator by the figure for Country of birth not stated, which should estimate the census non-response rate. All other percentages, such as the sum-product measure, increase, especially in places with a high non-response rate. Inner city areas remain among the least Christian, but they are no longer outliers. Outback areas also increase their scores significantly. All percentages on this page after the first table are the adjusted ones. Currently for the religion stats in Mappage you can turn this adjustment on or off to see what difference it makes.

Because I have decreased the denominator but I am still using the same multipliers for the sum-product, that is now an overestimate of total weekly church attendance, because those multipliers were generated without any adjustments. (I might re-scale it later).

Highs and lows

Looking at the sum-product score across the cities of Australia at SA3 level, there is one common feature: The inner city scores lowest (despite the adjustments I’ve mentioned) while any high-scoring areas are outer suburbs. In Sydney [4a], there is a large bible belt in the north west with a few SA3s at 11% or more. Mt Druitt is highest with 12%. At the low end are Eastern Suburbs North (6%) and Auburn (6.1%). In Melbourne [4b], most SA3s are in the 8-9.5% range, but there are several inner SA3s below 7%, including Yarra (5.6%) and Glen Eira (5.8%). At the high end is Maroondah (10.6%), Casey North (10.5%) and Keilor (10.3%), with all of the outer east above 9.5%.  In Brisbane [4c], Springwood-Kingston is highest with 12.2% and Brisbane Inner is lowest with 7.2%.

In Perth [4d], we don’t see the large differences that the larger cities have. Cottesloe-Claremont and Fremantle are the lowest two with 7.3%. Kalamunda is highest with 10% and much of the south east is high. Similarly in Adelaide [4e], apart from the outlier of Adelaide City (6.6%), the SA3s range from 7.9% (Unley) to 10.5% (Campbelltown). My original reason for getting into this topic was to see how different parts of Adelaide compared, but most of the big differences are in the three biggest cities.

A number of SA3s outside the major cities score over 11%. Lake Macquarie West (11.6%) in NSW;  Toowoomba (12.4%), Mudgeeraba-Tallebudgera (11.3%), and Darling Downs East (11.1%) in Queensland; Barkly (13.7%), East Arnhem (11.3%) and Alice Springs (11.1%) in NT. All of south east Tasmania is below 6.5%, with some of Hobart at low levels. Besides Tasmania, nowhere outside the five big cities scores below 7.5%.


The most obvious feature I have mentioned is the difference between inner and outer parts of a city. Looking for a census variable to describe how “urban” or not an area is, the first one I came upon is Dwelling structures. When we use the scatter plot to compare our sum-product score with the proportion of dwellings that are “Separate house” rather than semi-detached or unit/flat/apartment, we get significant correlation: 65% in Australia [5a], 73% in NSW [5b], 76% in Victoria [5c]. An area with 85% separate houses will score on average 1% lower than one with 60% separate houses. This difference has increased since 2006, with all high-density areas becoming less Christian except the Gold Coast and Darwin [5d]. There is even a 63% correlation with households having more than two motor vehicles.

When I started looking into this, I thought poorer areas would score less. However, the opposite is true [6]. In every state and territory, there is a 25-60% negative correlation with median family income. An area where the median family makes $2000/week will score on average 1% lower than where they earn $1000/week. However, this is mostly due to the inner/outer difference – inner city areas have the highest incomes. I also suspect that poor suburban areas are likely to have more nominalism than rich ones, so these measures will undercount the number of Christians in rich areas.

Looking a housing tenure types, there is correlation with the proportion of dwellings owned with a mortgage (over 50% in NSW and Victoria) and negative correlation with the proportion renting (60% or more in NSW, Victoria and SA). The correlation with public housing varies between states: -30% in Victoria and +27% in WA.

Looking at age profiles, there is 77% correlation with the proportion aged 5-19, and -43% with the proportion aged 25-34. On household composition, 51% correlation with single parent families and -63% with lone person or non-family-group households.

What next?

The main thing I want to do next is get hold of more church attendance data and change my multipliers accordingly. I might also set up different graphs in Mappage (eg to show time series) and test correlations with different census variables.  I hope to update this page as things develop.

1 Comment »

  1. Fantastic work – Thanks Eric!

    Comment by Darren — 23 June 2014 @ 11:15 pm

RSS feed for comments on this post. TrackBack URI

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Blog at