I will be giving a talk at a conference on the one-year anniversary of the marriage law postal survey at ANU in November, and for this talk I’ve put together a paper analysing the strongest demographic correlations which can help explain who voted which way in the marriage survey.
I’ve decided to put it up as a blog post, but it is a bit longer than my usual content. It includes a series of tables and charts showing how the yes vote correlates with various demographic data.
How equality was won: the demographics of the Yes and No voting coalitions at the 2017 marriage law postal survey
In the Australian Marriage Law Postal Survey of 2017, over 12.7 million Australians submitted forms expressing their opinion on changing Australia’s marriage law. 7.8 million people (61.6%) said yes, with 4.9 million people saying no.
Despite not being conducted as a plebiscite using standard Australian practices, the survey practically functioned as a public political vote of the people, the first such national vote since the twin constitutional referendums of 1999.
The electorate divided into two coalitions: just over three-fifths of the country who voted yes, and the remainder who voted no.
While the issue of marriage equality is likely now resolved, these coalitions are likely to be seen again in the future on other social issues, and it’s worthwhile examining who made up these coalitions.
Methodology and the ecological fallacy
I have examined the official returns of the Postal Survey, which provided a total vote for “yes”, “no”, and “invalid” for each of the 150 federal electorates used at the 2016 federal election. The official returns also include data on the number of people who turned out to vote amongst men and women and amongst each age bracket for each of the 150 federal electorates.
This is not a detailed level of information on how people voted. The official results of the 2016 federal election were broken down to the level of over 8000 local polling places across the country. Likewise the last major public vote (the republic referendum in 1999) also included published results at the polling place level.
I have compared this data to the results of the 2016 census at the same electorate levels. Unfortunately this means we only have 150 data points to examine, which will obscure demographic and voting variations within each electorate.
For this reason, I will only be looking at which demographic groups had a concentration which correlated with the yes or no vote. I will also be using some polling data to examine how age and gender interacted with the vote, and I have also compared the results of the postal survey to the results of the 2016 federal election. This is not the most sophisticated statistical analysis of the data available, but I think it reveals a lot.
Due to this data limitation, it is important to include a warning about the ecological fallacy. I will be able to say that electorates with a relatively high number of certain demographic groups had a relatively high yes vote, but this does not mean that those people necessarily voted yes, but it does suggest those demographics are more likely to vote yes, and it also suggests that people who live in the same communities as those demographic groups are more likely to have supported marriage equality.
I should also note that I will use the term “vote” to refer to how people filled out their surveys. I will be treating this survey as a political vote as it was originally intended. Despite using the systems of the Australian Bureau of Statistics I believe the survey’s content (as well as the campaigning around it) classify it as a political voting act, not a survey of opinions.
Key demographic groups in the yes coalition
The strongest correlation between the yes vote and any demographic group was with people who identified as having no religion, with a correlation of 0.8.
The no vote was clearly higher in more religious areas. Almost every major religion in Australia had a correlation with the no vote, with Christianity, Buddhism and some major Christian denominations having a negative correlation around 0.3. The religious group with the highest negative correlation was Islam (-0.56).
It’s worth noting that, while Islam had a high correlation, it only makes up about 2.8% of the population, so the ‘no’ vote included a lot of people who are not Islamic but live in similar areas, and also reflects that the no vote was a result of broad religiosity, not one religious group.
|Presbyterian and Reformed||-0.13||2.5%|
I have identified a series of other datapoints which are highly correlated with the yes vote (although not as strongly as ‘no religion’). These datapoints can be grouped into two clusters based on their own high correlations with each other.
The first group are datapoints which suggest socially progressive communities. There is a correlation of approximately 0.5 between the yes vote and the proportion of couples who are same-sex couples, and the proportion of people who have never married. These two datapoints are highly correlated with the Greens vote at the 2016 federal election, which itself has one of the strongest correlations with the ‘yes’ vote.
|Greens vote, 2016 federal election||0.65||10.2%|
None of these groups are big enough to explain the yes vote. But this data does suggest that the yes coalition was concentrated in places with more people who have never been married, and in places where LGBTI people are more welcomed and are able to openly live in couples.
Wealthy and educated
The final cluster reflects a series of demographic groups which reflect high levels of income and education.
|Median family income||0.51|
|Bachelors degree or above||0.37||22.0%|
|Women in the workforce||0.57||56.0%|
There is a clear trend of seats with higher median incomes producing higher yes votes. There is also a strong correlation with people who work in professional occupations, and the proportion of women who work.
There is also a significant correlation with the proportion of people who have a bachelors degree or greater, but it isn’t as strong as the above datapoints. There is also evidence of the yes vote doing better in places with greater internet access.
Age and gender
Examining data at the electorate level is not able to give us much information about how men and women may have voted differently, since there is not a great deal of variation in the gender and age balance between electorates. We fortunately have a lot of information from polls about how the vote varied amongst age groups and between men and women.
The trend from the polls is very consistent. The proportion of people voting yes declined amongst older groups, while women were more likely to vote yes and men were relatively more likely to vote no (although a majority of men appear to have voted yes).
These tables show the gender and age breakdowns from two polls chosen from the many conducted, although the trend was very consistent.
YouGov/Fifty Acres, October 12-16
Essential, October 23
Relationship between yes vote and turnout levels
During the campaign, polling quickly identified that there was a relationship between someone’s likelihood of voting and their voting intention – those planning to vote yes were more likely to vote. This could be seen in crosstabs on those two questions, but also in the demographic data: younger people and women were more likely to vote, and more likely to vote yes.
Essential, September 5
|Will definitely vote||74||58|
|Will probably vote||17||14|
|Will probably not vote||2||8|
|Will definitely not vote||<1||9|
|Not enrolled at current address||4||5|
This polling trend remained until later in the campaign, with this poll showing the difference in voting intention between those who had already voted and those who hadn’t:
Essential, October 4
|Already voted||Not yet voted|
This trend can be clearly seen in the final results. There was a 0.5 correlation between an electorate’s turnout level and its yes vote.
The difference in enthusiasm between yes and no voters can also be seen in the number of new enrolments in each electorate – there was a 0.62 correlation between the proportion of new enrolments in each electorate and the yes vote.
What about the multicultural effect?
I would expect you to be wondering, what about the effect of multiculturalism on the yes vote? It’s an understandable question: most of the electorates which voted no were in the multicultural western suburbs of Sydney (and some multicultural areas in Melbourne), and the dominant narrative in the days after the result pinned the blame on Australia’s more recent migrants.
Yet my analysis suggests this is a much less significant factor than religion in explaining how people voted.
I looked at a number of census datapoints which may suggest a multicultural community. There was some correlation with people who speak English in the home (0.39 correlation with yes vote), but there’s pretty much no correlation with the proportion born in Australia.
I’m not going to say there is no correlation, but it’s worth pointing out that the non-English-speaking datapoint is only half as correlated as whether someone has a religion.
When you look at correlation for different ancestry groups, you find a big difference in correlations. There is pretty much no correlation with the yes vote for electorates with large Chinese and Korean populations, and the negative correlation for the Indian ancestry group is relatively small. There is a larger correlation for some middle eastern communities, which I suspect can be explained by higher religiosity levels in areas with large communities from the middle east.
Interestingly, multiculturalism looks much more important if you look just at the result in New South Wales. Twelve out of seventeen seats with a majority no vote were in New South Wales, all of them in a contiguous areas in western and southern Sydney: all of them seats which are much more multicultural than the national average.
The following chart shows that NSW clearly underperformed, providing most of the worst-performing seats amongst those electorates with lower english-speaking proportions. But it is also the case that the five seats with the lowest proportion of people speaking English are in Western Sydney – you don’t get the same concentrations in other big cities.
Very multicultural communities were concentrated amongst a handful of the most strongly no-voting electorates in our country’s biggest city, but the level of cultural diversity does not have any explanatory power outside of this region. Indeed, the predictive value of non-English speaking drops from 0.39 to 0.11 (basically nothing) if you exclude the seven Western Sydney seats with the biggest no vote.
Over 60% of Australians voted yes in the postal survey. This is far too large and broad a group of Australians to be pigeonholed into a handful of demographic groups, most of which make up less than half of the population.
But the trends are clear. The yes vote was won most strongly amongst communities with lower levels of religiosity, higher levels of wealth and education, and socially progressive communities with relatively low levels of marriage.
And while a lot of focus has been on multicultural communities in Western Sydney for voting no, cultural diversity does not explain any of the variation in the yes vote outside of this one region. The strongest evidence suggests it was religious voters (both white and non-white) who underpinned the no coalition.
Yes voting intention
- Essential poll, October 4, Do you intend to vote yes or no? (Based on those who have not yet voted)
- Essential poll, October 4, Did you answer yes or no to the question “Should the law be changed to allow same-sex couples to marry? (Based on those who have already voted)
- YouGov/Fifty Acres, October 12-16, Voting intention (for people who have voted and haven’t voted)
- Essential poll, October 23, Did you answer yes or no to the question “Should the law be changed to allow same-sex couples to marry? (Based on those who have already voted)
Likelihood of voting
- Essential poll, September 5, How likely are you to vote in the national postal vote on same-sex marriage?
Image at the top is of marriage equality rally in Parramatta, 29 October 2017.