Nerd alert; this post is chock-full of graphs. Plus a few “hypotheses”, just to keep things exciting. And a dog, because I like dogs.
For those of you who are new to the data game but want to participate in the nerdy excitement, let me first explain the rules. The game starts when an economist, such as myself, formulates a “hypothesis”. In doing so we’re basically statistical crystal-ball gazing. Not only is this fun, but it’s actually essential.
It’s essential because formulating a hypothesis sharpens/hones subsequent analyses. It’s also essential because of what it reveals about our own internal biases. Economists have long recognised that we’re at least as biased as anyone else (and typically more selfish) – so it just seems kosher to clearly state our biases from the outset. That way other people know what game we’re playing.
I’d encourage you to play the economics game sometime, with just one word of caution: Be prepared to be wrong. I’m 100% certain, for example, that I will – at some point in the near future – be wrong. Keep that in mind whenever you are tempted to believe your own EBS (economic BS). I might also add, however, that economists don’t care much for being right, we just try and make damned sure we’re less wrong than the engineers.
Professional happiness is, after all, largely a relative frame of mind.
The primary hypothesis (it’s not really “mine” – we’re all standing on the shoulders of nerdy giants here) is something that has been articulated in previous posts here and here. Kent recently talked about it again here, where he also came up with the crazy notion that I should write a follow-up post. Thanks for that hospital pass. But other people have also asked for it, so I finally pulled finger and cranked this out. Unfortunately the cacophony of requests for follow-up posts misses an important issue: Not much additional data has been released since my last posts on the topic. Nonetheless, even in covering well-trodden ground I did find some interesting new nuggets.
The hypothesis is this: The demand for vehicle travel in NZ (when measured in terms of vehicle-kilometres travelled per capita) is falling in response to powerful wider forces, including:
- Demographic trends, such as an ageing population (which has both age-related and income-related effects)
- Socio-economic preferences, such as reduced attachment to private vehicles or increased awareness of health/environmental impacts of driving
- Technological developments, such as smart phones, PT journey planners, tablets (which help to demystify public transport and lower the perceived costs of in-vehicle time)
- Trends in transport costs, most notably sustained higher fuel prices but also reducing costs of air travel (Air New Zealand recently launched $29 late night fares between AKL-WGN)
Stylized facts support this hypothesis. In the graph below, for example, I’ve plotted VKT per capita p.a. in New Zealand over the last decade (source). You can see there was a trend towards higher VKT per capita until circa 2004, after which the trend turned negative. Since it’s peak VKT per capita has fallen by about 6% (NB: I’ve base-lined results to 2001 levels, so the graph measures the % change from that year onwards. Doing so makes it easier to identify and compare trends, which will be useful later in this post).
The good news is that this graph suggests I’m not a complete moron. But nor does this mean my hypothesis is necessarily correct.
An alternative hypothesis, which has been advanced by the Government, is that the drop in VKT per capita that has occurred since 2004 is a temporary aberration caused by post-GFC economic malaise (which sounds suspiciously like a budget form of mayonnaise). Well, let’s investigate what I call the “budget mayonnaise” hypothesis by adding (indexed) real GDP per capita to the same graph (source).
Hmm. On the basis of this evidence it’s fair to say that the Government is more wrong than me: Budget mayonnaise does not seem to explain VKT per capita trends very well. The data actually contradicts their hypothesis; unless they’re going to suggest that drivers started preparing for the 2008 GFC way back in 2004. Now it is true that the GFC has a negative impact on GDP per capita, but this impact was relatively slight and has since been more than wiped out. In fact, since 2004 – which was the peak in per capita VKT – our GDP per capita appears to have increased by about 12%, whereas per capita VKT fell by about 6%.
At this point, ardent purveyors of the budget mayonnaise hypothesis might suggest GDP per capita is not the only indicator of economic activity, and dog-yarned it they’d be right. Wages, in particular, tend to “lag” movements in GDP, such that it is possible that post-GFCNZ is experiencing suppressed wages. But why bother asking when we can get busy answering – in the figure below I’ve added inflation-adjusted hourly earnings (source).
Here we see earnings rising steadily from 2001 to 2009, since which time they have fallen. This drop probably reflects the lagged effects of the GFC plus the Christchurch Earthquakes. I should say that 2012 data shows earnings rebounding back up to 2009 levels, which is good news. But overall these indicators do not provide much evidence to support the view that a slowdown in economic activity is primarily responsible for declining VKT per capita. It seems fair to conclude that the budget mayonnaise hypothesis does not cut the mustard (how does one actually cut mustard?).
That’s not to say economic activity in general does not impact on VKT per capita; I definitely think it does. And I certainly expect that as the economy gathers momentum (as it seems to be) VKT per capita should “rebound” somewhat. Whether this rebound is sufficient to counteract factors that are causing it to decline is hard to say. My hunch is that if fuel prices stay low then we may see some VKT growth, but that’s not really something the Government can point to and say “we told you so”; that just strikes me as getting lucky.
So for now at least, I’m sticking to my story – even if I look forward to the MoT updating their vehicle fleet spreadsheet with 2012 VKT data so that I can finally be proven wrong, and in turn get to start the game again with a new hypothesis (plenty more where that came from).
But I’m also sticking to my story because, in writing this post, I uncovered another little piece of data that seems to support my hypothesis. But it does so in a somewhat unusual way, in that it suggests that VKT per capita should have actually grown in the last decade. The source of this indicator is this delicious data set from Statistics NZ,which presents New Zealand’s population from 1991 to the present. You might think that’s not particularly exciting or novel. Think again – because in this data set Statistics NZ has split the population by their precise age at the time. It’s rather useful I think.
Using this data you can estimate the number of people of driving age (which I’ve defined to be 16-70 years). You can then calculate the ratio of people of driving age per capita, which measures the proportion of the population who are of an age where they could get a drivers license, if they were so inclined. I’ve plotted this ratio below for the period from 1992 to 2012.
Holy bandages. What this graph shows (if the Statistics NZ data and my calculations are correct) is that the proportion of people of driving age in New Zealand is now at the highest level it’s been for two decades. Moreover, most of the growth in this ratio has occurred within the last decade. The very same decade that has seen a fairly significant decline in VKT per capita.
Thus in the same decade when the proportion of people of driving age appears to have increased, we have witnessed declining VKT per capita.
Fairly interesting stuff. And well worth chucking into a regression model, if we have more than 10 years of data.
Where this ratio may head in the future I just don’t know, although Statistics NZ might be able to tell us. Perhaps I’m over playing it’s importance – after all the ratio only changes by 2.5% over a period of ten years (note the graph’s truncated vertical axis). But I would have intuitively thought (here we go again with the same old EBS) that the proportion of people of driving age would have a relatively large impact on VKT per capita. Actually, in a situation where young people behave exactly like their parents you might even expect an elasticity approaching 100% (i.e. a 1% increase in the proportion of people of driving age would cause a 1% increase in VKT per capita).
Of course, the proportion of people of driving age is only one part of the VKT equation. One useful (albeit incomplete) way to visualise the VKT equation is to consider a Russian doll of overlapping circles, where each circle captures related but distinct demographic and socio-economic variables, which ultimately combine to determine VKT per capita in any given year, like I’ve shown below.
The outer-most circle is population, which is important only in a “multiplier” sense. That is, we can forecast total travel demands by multiplying VKT per capita by NZ’s total population. What the previous graph showed is that the lilac-coloured circle has increased in size relative to the outer circle, i.e. a greater proportion of the total population are now of driving age per capita.
We also know from our previous analyses that the inner-most (orange) circle labelled “DRIVE” (i.e. VKT per capita) is currently declining. These two results thus suggest that the reduction in VKT per capita is most likely to have arisen in response to either 1) a reduction in proportion of people with drivers license per capita (green) and/or 2 ) a reduction in vehicle ownership per capita (blue circle).
The graph below shows the latter (per capita vehicle ownership) superimposed on top of the previous VKT trend.
So it has declined, by not by much. And what is most interesting about this graph is that the peak in vehicle ownership occurs a couple of years after the peak in VKT per capita. Does this suggest that travel demands are actually the egg and vehicle ownership is the chicken? Could it be that NZers are choosing to reduce their VKT first and only subsequently reducing the number of cars they own? Or maybe it’s both – maybe lower VKT results in lower vehicle ownership, and lower vehicle ownership results in lower VKT. I should say that 2012 data shows a small rise in vehicle ownership per capita back to 2010 levels, but it’s still down on the 2005 peak.
The final piece of my little diagram is the proportion of the population with drivers’ licenses. I don’t have this data on hand, although I’m sure it’s out there (please point me to it if you know of good data sets). Anyway, I think that’s quite enough statistics (and lies) for one day, so let’s finish with something that is 100% certain: Puppies that try to chew tennis balls even when lying down are 100% adorable.