Descriptive Analysis

The Stimulating A Sunburnt People map raises some interesting questions. It shows the amount of federal stimulus funding spent on community infrastructure and road and rail projects per Federal Electoral Division shaded by the party affiliation of the federal House of Representatives member for that division. Exploring the maps, it's easy to wonder: has funding been divided evenly by state? How has it been split between metropolitan and rural areas? Do the size of an electorate or its population density make a difference to the funding it receives, independently of whether it's a metropolitan or rural district? Does the political party affiliation of an electorate's representative influence the amount of funding they receive, after all these other influence have been accounted for?

Some of these questions can be answered by analysing the aggregate data. First, we present the funding allocations by state and party, and then describe the results of a more detailed analysis. As you read, please keep in mind that we only have data on community infrastructure and road and rail projects, it is quite possible that ferderal spending on these projects is not representative of how the remainder of the stimulus package has been and will be allocated.

Stimulus Spending Per State

State Number of projects Total value Value per capita Percentage of ALP senators NSW 1461 $1,468,185,746 $224.53 50.0% VIC 981 $1,699,884,591 $345.08 41.7% QLD 557 $2,451,946,764 $630.04 41.7% WA 529 $214,667,869 $109.91 33.3% SA 484 $540,937,461 $357.83 41.7% TAS 192 $197,946,304 $416.20 41.7% NT 56 $16,132,094 $84.46 50.0% ACT 10 $3,791,700 $11.73 50.0%

Stimulus Spending Per Party

Party Number of seats Percentage of seats Number of projects Value of projects Percentage of value Ratio of Value to Seat percentages Australian Labor Party 83 55.33% 1416 $4,534,542,218 68.78% 1.24 Liberal Party of Australia 53 35.33% 947 $1,743,087,229 26.43% 0.75 The Nationals 9 6.00% 595 $235,759,844 3.58% 0.60 Independent 3 2.00% 300 $76,197,440 1.56% 0.78 Vacant 2 1.33% 12 $3,905,798 0.06% 0.05

NB: The government did not assign a value to the 292 individual projects that are part of the $150M Boom Gate Project. We assigned a value of $150M/292 to each of those projects for our calculations.

Inferential Analysis Results

A few features of the map stand out quite clearly - metropolitan areas seem to be getting very different levels of funding than rural and provincial areas. The bar graphs above suggest that ALP seats are getting a larger share of the funding than non-ALP seats. Could these facts be related? The ALP tends to hold more metropolitan than rural seats. Rural electorates are usually much larger than metropolitan ones and may require projects on a much larger and more expensive scale. Commonwealth Electoral Divisions are designed to contain roughly the same number of people, though this varies a little, as a result population density varies dramatically - metropolitan electorates are much more densely populated than rural ones. Dense city-living may require a very different level of investment in community and road and rail infrastructure than country-living does.

One well-established and relatively simple technique commonly used by scientists and economists for disentangling such competing influences is called Ordinary Least Squares Regression. Using this technique we found that projects in metropolitan areas cost, on average, more than projects in rural and provincial areas but that this difference is balanced out by the number of projects in each area so that city-folk and country-folk receive almost exactly the same stimulus funding per-capita. Interestingly, no matter what measure or scale we consider nor what demographic variables we control for, ALP electorates always seem to get quite a bit more than non-ALP electorates, however that the data available to us can't conclusively rule out the possibility that ALP electorates just got lucky. Please remember that these conclusions only pertain to spending on community infrastructure and rail and roads. Given the data available to us, we have no way to evaluate how representative they are of other Economic Stimulus Plan spending, for instance on housing, education or solar hot water, which may dispropotionately favour non-ALP electorates.

We found that community infrastructure and rail and roads projects in rural areas were on average funded for $2,695,315 less than projects in metropolitan areas. Even after this city-country difference had been factored out, projects in ALP seats were still funded for, on average, $1,606,089 more than projects in non-ALP seats. Given the how diverse the levels of funding were per project, we estimate that there's about a 16% chance that projects in ALP seats received this much more funding just by the simple luck-of-the-draw.

Per-capita, the average non-ALP electoral division received $280. There was almost no difference per-capita between metropolitan an rural areas, however ALP electoral divisions received, on average, $345 more, giving them an average of $625 per capita. However we estimate a 33% chance that this advantage to ALP electorates was simply the luck-of-the-draw.

Per-square-kilometer the ALP electoral divisions received, on average, $266,523 more than non-ALP divisions, even when we statistically accounted to the fact that on average metropolitan areas received $261,792 more than rural areas. We estimate about a 22% chance that these differences were just the luck-of-the-draw.

A more detailed description of this analysis and simplified regression tables are provided below. Further details can be obtained by contacting Maciek Chudek.

Analysis Details

To get a more detailed sense of what this regression analysis is telling us, consider first very simple model where we look at the just the influence of whether a project occurred in an ALP-held Commonwealth Electoral Division on how much funding the program received.

We can see from this table that the average project received about $1,762,640 in funding and that projects in ALP seats received on average $2,196,513 more than projects in non-ALP areas (little number crunching also reveals that projects in non-ALP seats were funded, on average, for $664,383). The final column tells us how likely it is that we'd see this large a difference in funding allocation, had the funds been allocated without the political affiliation of the electoral divisions having any influence. Usually scientists consider values of 5% or less to to be good evidence that a variable is really having an influence.

Of course, this extra money might have been allocated to ALP seats just because of their size, population-density or whether they're metropolitan or rural. To evaluate this possibility, we can include these variables in the model simultaneously*. This shows us that:

The bulk of what seemed to be a difference in allocation by the party an electorate elected is really being caused by differences in whether electorates are in rural or metropolitan areas.Rural areas received about $2.5m less funding per-project than did metropolitan areas. Notice that even after this is accounted for, ALP seats still received more funding on average, however there's now an estimated 16% chance this was just good luck.

So far we have only considered the amount of money allocated per-project. Perhaps metropolitan and ALP seats had fewer more expensive projects, but rural non-ALP seats had more, less-expensive projects.

We can evaluate this by looking at the total funding allocated per electoral division (which you can explore on our map). First let's consider the raw funding*:

Now we see the differences between metropolitan and country areas all but disappear. The higher cost of projects in metropolitan electoral divisions seems to be offset by more projects in rural areas until the funding to both is almost identical. However at this level we still see ALP seats getting a larger share of funding, on average, than non-ALP seats, though there's a much larger chance that this was just good luck.

It's also interesting to consider the amount of funding an electoral division receives per-square kilometer*:

Here we see the same pattern as we did per-project, ALP divisions receive more funding per square kilometer, even after we factor out the fact that country areas receive less. However both these effects have about an estimated 22% chance of just being a result of blind luck.

Finally it's worth considering the per-capita funding differences between electorates. Federal Electoral Divisions area designed to have roughly the same number of people, however as populations grow and move this can quickly become quite a rough estimation. It may well be the case that city electorates have on average more people and thus need greater funding. We can investigate this by investigating per-capita funding*:

Again we see the effect of being in a rural rather than metropolitan area almost disappearing - city-folk and country-folk have gotten a roughly equal share of the pie. Still, even after this is accounted for, people in divisions which elected and ALP representative received on average $344.50 more than people in divisions represented by another party. However, given the information we have available, there's a reasonable chance that this difference between ALP and non-ALP electorates occurred by chance.

Together, these models paint an interesting picture. Metropolitan electorates recieved considerably more funding for community infrastructure and road and rail projects than did rural electorates, nonetheless this seems to have balanced out very evenly per-capita. Yet no matter what scale we consider the data at, nor what demographic variables we control for, ALP electorates seem to receive more on average than non-ALP electorates, not not enough more that a reasonable argument couldn't be made that this was just blind luck.

Again, please recall that these results apply only to spending on community infrastructure and road and rail projects. Given the data available to us, we are not in able to make any estimations of how representative these trends are of stimulus or federal government spending in general.

Footnotes

*A note to the statistically-minded: For all models including continuous predictors of area, population and density did not yield qualitatively different results of improve the fit of the model, nor did further discriminating metro/country into inner and outer metro and rural and provincial areas. The simplest qualitative predictors are presented here for ease of interpretation.