In his new book, George Megalogenis “argues that Labor lost in 96 because it lost
the middle and upper income voters who’d given support to Hawke and
Keating, and that the electorates with the lowest taxable incomes had
supported the Liberals by a majority in previous elections Labor won”, writes Mark Bahnisch at Larvatus Prodeo.

Barely a month goes by without some journalist or another trying to
work out what individuals are doing by how their electorates are
swinging. The problem with this is that there are only 150 electorates,
so trying to guess demographic patterns becomes very tricky very fast.
The result is that journalists from Megalogenis to Ramsey end up committing the ecological fallacy. As Wikipedia defines it:

The ecological fallacy is a widely recognised error in
the interpretation of statistical data, whereby inferences about the
nature of individuals are based solely upon aggregate statistics
collected for the group to which those individuals belong. This fallacy
assumes that all members of a group exhibit characteristics of the
group at large.

Simply put, if you want to know how poor people vote, then looking
at the voting patterns of electorates with low average incomes is an
extremely rough proxy. A better approach is to use the Australian
Election Survey, which asks people how they voted, and what their
demographic characteristics are.

Now this sounds pretty straightforward. If you want to know how poor
people voted, ask a bunch of poor people how they voted. So why don’t
smart press gallery journalists with six-figure salaries and research
assistance do it?

I think the answer is that using the electorate-level data is just too tempting. Glance at the Parliamentary Library and AEC
websites, and you’ll get some nice tables on voting and average income.
By contrast, if you want to run some cross-tabulations from the
Australian Election Study, you have to go to the clunky Nesstar interface, or read drearypapers by academics. The only consolation is that you might actually get the right answer.