How Checkout Chicks Smashed Howard.

Sometimes when you’re playing with electorate level demographic data, you get the occasional WTF! moment – some relationship between a demographic variable and party vote share pops up that is so big it makes you take a deep breath and a second and third look. Unusually large, strong relationships are rare with this sort of aggregated data and on the few occasions they do occur, it’s usually some sort of ecological fallacy writ large that is responsible for it.

But pottering around with the census data this morning, something jumped out and slapped me – a relationship between the ALP two party preferred swing at the last election and a certain variable was so large it literally screwed the constant up in the regression equations.

The variable comes from the answers to this question in the Census:

But specifically, the geographical distribution of the people that marked the “Retailing (incl. Take-aways)” box. You can read more about the technical ins and outs of the question at the ABS.

I pulled out of the census the percentage of the total population of each electorate that marked “retail” as an answer and compared it to the ALP swing by electorate at the last election. The results are incredible.

Remembering back to last year, the retail sector was at the pointiest of pointy ends of the Workchoices debate – it was where the data suggested that the biggest losers of the Workchoices policy were employed. State government research, Union backed research, academic research and eventually the governments own numbers all told the same story and received wide media coverage. To give a few examples:

NEW STUDY SHOWS WORKCHOICES CUTS PAY IN RETAIL AND HOSPITALITY INDUSTRIES.

RETAIL AND HOSPITALITY: KEY CASUALTIES OF WORKCHOICES

WorkChoices Hurting Retail Workers

I’m sure you can all remember the general gist. Apparently the retail workers did when they got into the ballot booths.

First up, the actual data itself. The numbers might at first glance appear small as we are only using the percentage of the total electorate that market retail on the answer to that question above, so all of those people not in the labour force water down the headline numbers, and many of the folks actually working in a retail area of a mostly non-retail business would have answered something different. But don’t let the small proportions fool you.

If we use a histogram to describe the data across all 150 electorates we get:

As we can see, no electorate passes the 1% of the total population mark answering this question as “Retail”, so what we are effectively looking at here is a pretty tight definition of retail employment.

If we run this as a scatter plot against the actual ALP TPP vote achieved at the last election, by electorate, we can see that retail employment wasn’t particularly skewed to the size, or the stock of the Labor vote.

There is no statistically significant relationship there. But what is interesting about this data is the way it partially relates to education. If we run two scatterplots of retail against the proportion of each electorate that has a year 10 education or less and also run a linear fit through one, and a locally weighted polynomial regression fit though the other we get:

Over the entire spectrum of values, there is no linear relationship between the average increase in the percentage of each electorate that works primarily in retail and the proportion of the electorate with a year 10 education or less, but if we look beyond the average, retail employment increases as the number of folks with a yr10 education or less increases up until about about 35% point, after which it starts to walk hand in hand with a reduction in the proportion of the electorate that works primarily in retail.

The other thing worth noting goes back to how we saw the growth in the population aged 65+ over the period 2006/7 correlating with an increase in the ALP swing by electorate. But we may have found out a partial explanation for this anomaly since the 65+ age cohort growth correlates strongly with this retail variable:

But now for the actually point of all this. This is how the correlation between retail and the ALP TPP Swing looks across all 150 electorates:

As a linear regression, it is huge for this type of data – and the nature of the relationship screws up our constant somewhat (which is not a bad thing – it’s just an unusual thing for this sort of data)

A 1% increase in the proportion of the electorate that answered “Retail” to that Census question walked hand in hand with, on average, a 12% increase in the ALP TPP swing. But since we’re dealing with such small proportions on the retail side here it’s probably best explained as a 0.1% increase in the proportion of the electorate that answered “retail” to that Census question walked hand in hand with, on average, a 1.2% increase in the swing to the ALP. This retail variable alone explains statistically 15% of the variation in the ALP swing by electorate.

But let’s look at how it played out in just the metro seats.

In the metro seats the relationship was larger and stronger and statistically explained even more of the variation of the swing. This time a 0.1% increase in the “Retail” answer walked hand in hand with, on average, a 1.6% increase in the ALP Swing by electorate to explain 27% of the variation of the ALP swing.

If we now throw the growth in the 0-4yr age cohort over the 2006/7 period into the mix in the metro seats – which we found previously had a strong and significant relationship with the ALP swing (although in an opposite direction to the retail variable), we get.

This suggests that some group of things that correlate with rugrat growth in metro seats combined with retail employment numbers in metro seats explains 38% of the variation in the ALP swing in those metro seats.

For every 1% increase in the rugrat population over 2006/7 (holding retail constant) there was on average approximately a one half of one percent reduction in the ALP Swing, and for every 0.1% increase in that retail answer (holding rugrat growth constant) there was on average an approximate 1.4% increase in the ALP swing for metro seats.

Ordinarily this would be hard to believe – but the power of Workchoices might have been a little bit more powerful than even the union movement is suggesting when it came to getting the ALP elected. It might be worth some deeper research.