Every half-decade, Census data gives us an interesting and detailed insight into how New Zealanders are travelling. Back in August, the Ministry of Transport published a comprehensive analysis of journey to work patterns in Auckland (ably summarised by Matt here).

Here’s one of the key maps from the report. It shows average distance travelled to work for all suburbs in Auckland. Blue means shorter trips, red means longer trips. As you can see, average commutes get a lot longer if you live further from the city centre:

Trip Length residential 2013(I’ve used the same data to take a look at issues like housing and transport costs and greenhouse gas emissions from commutes in NZ’s large cities. It’s definitely a rich source of insight into how we live.)

The Census journey to work data presents a conundrum. Auckland is not a monocentric city in which all employment is concentrated in the centre. It is in fact highly polycentric, with employment dispersed throughout a number of locations. The map below, which I put together quickly using Statistics NZ’s Business Demography employment data, shows this. There are certainly many jobs in the city centre – around 15% of the total – but employment is spread around the entire Auckland region.

Auckland_employment_by_AU_2014

Given this, why aren’t people in outlying areas simply commuting to the nearest jobs, and skipping the long average commutes across town? Why aren’t the residents of Browns Bay commuting to Albany, the residents of Glen Eden to Henderson, and Howickians to East Tamaki? Auckland’s employment has long since decentralised – so why haven’t our travel patterns decentralised as well?

To answer these questions, we must consider the dynamics of urban labour markets. Here’s an illuminating graphic from Alain Bertaud’s recent talk in Auckland, which I reviewed here. It shows four different models for urban labour market, ranging from a totally monocentric city (all jobs in the centre) to a totally dispersed city (all jobs randomly dispersed). Auckland is clearly what Bertaud calls a “composite” city. It has a strong and growing city centre, but also a lot of jobs spread around other metropolitan centres, industrial parks, local shops, etc.

In a composite city, people do not simply commute to the nearest offices – they will actually travel to jobs all throughout the city. The “urban village” idyll simply doesn’t happen in real life. There are three big reasons for this:

  • First and foremost, labour markets are dynamic. Even if people start out working near where they live, this happy state of affairs doesn’t necessarily continue. Companies go out of business, workers get offered better jobs elsewhere, and people change careers. This happened to me earlier this year – a job change saw me swap my short commute from Mount Eden to the city centre for a longer commute to Takapuna.
  • Second, most households include multiple workers, who may have jobs in very different places. If you’re a baggage handler at the airport married to an accountant who works in Newmarket, it’s not going to be possible to live anywhere that offers you both a five-minute commute.
  • Third, people don’t necessarily want to live right next to their jobs. While commute costs are an important determinant of household location choices, we also consider a range of other factors, such as proximity to beaches and parks, school zoning, the location of family and friends, and so on and so forth.

Bertaud urban structure graph

Because labour markets are dynamic and people’s location choices are influenced by a range of factors, average commute distances tend to follow the location of the average job in the city. In other words, if you live in a neighbourhood that is ten kilometres away from the average job, you’d expect your neighbours to commute ten kilometres, on average. Some of your neighbours will have shorter commutes to local jobs, while others will travel longer distances to jobs on the far side of the city.

With that in mind, I’ve calculated the weighted average location of jobs in the Auckland region using Statistics NZ’s Business Demography employment data for 19 high-level industries. (Without going into the details, you can think about the method as follows: Let’s say that one individual industry has 200 jobs in Albany and 100 jobs in Takapuna. Then the weighted average job would be located two-thirds of the way from Takapuna to Albany. That’s what I did, except I was working with data on over 400 Auckland suburbs.)

The following map plots a centre point for each of the 19 industries. It shows that the average job is located in the Auckland isthmus. The average job in “blue collar” industries like manufacturing and warehousing tend to be located much further south – a result of the concentration of those industries in places like Mount Wellington, East Tamaki, and near the Auckland Airport. “White collar” industries like finance and insurance and professional services, on the other hand, are much further north, as they tend to be more centralised in the city centre and, to a lesser extent, in metropolitan centres like Newmarket and Takapuna.

If we look back to the first map, from the Ministry of Transport’s analysis of Census journey to work data, we can see that the geographic centres of Auckland industries fit within the blue swathe of relatively low average commute distances down the middle of the isthmus.

Auckland_employment_centre_industry_2014

In other words, centrality still matters even in a decentralised city! In a dynamic labour market, it is beneficial to live near the average job because it will tend to minimise expected commute distances over time as you change between jobs. That’s one of the reasons why prices are so high in the most central areas of Auckland: people seem to be paying a premium to be closer to the average job.

Of course, the data in the last map also shows that workers with different skills may have different optimal locations. If you expect to work in a “blue collar” industry, living further south might be a better strategy for minimising your expected commute distances. On the other hand, living further north might be better if you’re expecting to work in office jobs. However, labour markets are dynamic in another way as well – people may retrain or change industries throughout the course of their lives, and children may aspire to different professions than their parents. If that’s the case, living closer to the centre still offers more flexibility.

Share this

24 comments

  1. Until the Christchurch City centre returns at strength our second city is most like B on Bertaud’s classification and the resultant travel outcomes are abundantly clear: traffic congestion of a city many times it’s size, land use dominated by parking, poor pedestrian experience, place quality subordinated to vehicle through movement, weak PT uptake [though services have been directly affected by the quakes too- and will now hopefully improve with the New Network].

    In contrast Auckland’s strengthening centre has enabled improvements in pedestrian quality, intensification of land use and economic activity, a drop in private vehicle use on city streets, a dramatic jump in PT uptake. AKL is clearly composite in form, like most cities, but has the reversed last centuries trend for decreased centralisation and is moving back towards having a stronger core. Is becoming more like an older city form than a New World one.

    This is a good thing as it’s clear that B the highly distributed city model is both expensive and inefficient and leads to lower quality built environment for its citizens.

  2. Great piece of analysis. The number of times i read in places like the herald that we can keep sprawling over the countryside and it wont hurt congestion because jobs are dispersed in auckland and therefore people work locally….

    1. Ha, yes! It’s like, haven’t these people realised that their decentralised utopia has _already_ happened, and failed spectacularly to solve Auckland’s problems with road congestion?

  3. If we implemented your 2030 network with 40% buses even with a class b network dedicated lane with bus priority plus only pick up at main centres
    40% frequent network, 20% gathering on fringes it wouldn’t really matter too much where you.live and work. Then keep converting the class b network to class a ie rail or bus equivalent like your plans to get up to 5 star. PT now can compete with car freedom for workers and school pickup/ dropoff. Now widescale and fast. For $10 day pass now reasonable also like calgary fare and they have a higher fare box recovery than we do.

    1. Fully supportive of a more compact city and better options cbd. I just think our disfunctional road network shouldn’t preclude a longer commute for any.mode. Everything needs to flow and firmly believe we could turn this around overnight if Len and David wanted to with absolute vigour. Frankly we shouldn’t open any housing not close to rapid transit until at least up to Class B bus network on core skeleton to the housing area and congestion on way to be kicked by I eat cars for breakfast Mr Bus with own lane and signal priority, double teaming with train the terminator and bus on steroids North Shore Expressway.

      1. All Len and David need to do is say to the road maintenance team in each zone and signal team at Smales Farm. Give bus one lane via bus symbols and make them go fast with no delay. PT get buses under dayworks at least the 400 required on the rapid network 2030 plan. and quickly reconfigure bus stops for multi transfer and all bus stops off rapid network. Implement Frequent network now with complimentary outer collection.
        Bus come out fighting without hands and legs tied up over entire roading network. Congestion on way to being defeated even without property rates on carparks nuclear arsenal which you can choose location and how many warheads or just enough to give PT a nudge so roads and motorways not stressed and no unnecessary damage to economy. If you have to work other side of town at least not an exponential wait.

  4. This analysis is very good. The composite model is relatively common. Certainly in the US. The dispersal of jobs followed the same path as residential: new tracts of cheap(er) land, the promise of population growth to support service and retail trade, and most important, the motorway system. (In the US, anyway, sprawl is directly correlated with motorway growth, and is the strongest predictor of sprawl.) Even though employers wanted to move out of the centre, they still wanted to be accessible to it because it is the business hub of the region. An exception would be warehousing/distribution that located near interchanges to minimise exposure to local traffic congestion, narrow streets, and allow larger loading and marshaling areas.

    This dispersal came about over a few decades. One of the results was the reverse commute. It could be argued that the motorway system, or transportation system in general, was not designed for this which explains why so many road widenings happen in outlying areas. Not sure of the implications of that, if any, except that suburban road-building often comes at the expense of central area road maintenance and expansion (which is also constrained by available land on which to expand).

  5. A small request – can we please please please use the employee *density* in the second map instead of a raw count for symbolizing where people work. The problem is the map heavily skews data between block sizes (which are inconsistently related to resident population). The net result is the map doesn’t actually give you nearly as much insight into work patterns as you think it does.

      1. Hi Hamish – that’s a useful bit of feedback. I did in fact think about mapping the data in terms of density but decided not to for two reasons:
        1. I wanted to convey a rough sense of how many total jobs were in each area (or accessible to it). Density figures are a bit less intuitive than gross employment figures.
        2. The density map isn’t _that_ different than the employment totals map. The main areas of difference seem to be in the Auckland Airport and Wiri area units, which are physically very large. However, they also have a lot of jobs, which I thought was important to communicate.

        Also, Stu – I like your suggestion about calculating effective job density. In fact, I was already planning on doing this for a future post!

    1. problem with density is that absolute numbers are relevant when discussing spatial concentration more generally, i.e. beyond an area unit …

      so why not just map centrality itself? E.g. weighted sum of employment in an area unit plus employment in surrounding area units where the latter is factored down by an exponentlal decay function. Helps mitigate some of the arbitrary boundary effects …

      1. No, the absolute numbers are meaningless without accounting for the size of the area unit. The AU size (area) is roughly related to the size (number) of the normally resident population. If anything, the map tells you a little bit about the relationship between normally resident population vs employees, with a huge error bar because the resident population is not equal for all AUs and you can’t actually use them as an accurate proxy for equal population areas.

        If you want to look at centrality, it would be more useful to break it down to meshblocks and start running some cluster analysis over the results.

          1. Sailor boy, I’m pretty confident on my expertise on the subject ;-). Please read what I wrote and rebut the argument I’m making, rather than assuming I’m not experienced in the field.

          2. . The map was intended to show absolutejob numbers in each area of a region that the audience is extremely familiar with. The map does a good job of transferring the desired info to the desired audience. If it were done for a herald article then towers in monotone caus would work better. But since the absolute numbers are the message I struggle to see any audience where density would be appropriate.

            I slso note that you have made no justification to why density would be better

          3. tldr;

            The justification for density being better is it is actually comparable between area units of different sizes, unlike the absolute value, which is not comparable. Using the absolute value leads to misleading representation of geographic patterns in the map.

            The size of the geographic units used are not uniform – they’re highly variable. (the size of the units are loosely based on usually resident population, but that isn’t important for this discussion – it just means that the size of the geographic units is independent of – or at least, very complexly related to – the thing you’re actually trying to tell people about).

            The map is supposed to tell you something useful about the geographic distribution of employees between area units. Since we know the count of employees is a function of the area, we need to remove the effect of the area if we’re going to provide something comparable between units. We can do this by dividing the total number of employees by the area, which produces the density. Or as Stuart put it, the spatial concentration of employees.

            The density is actually comparable between area units, unlike the absolute value, which is not directly comparable.

            By the way: if your problem is that the units are not friendly to the audience, that’s a completely different question. You could break the density into quartiles to show the ‘most dense’ vs ‘least dense’ areas, or you could produce equal-area blocks that you can use to calculate an absolute number back from the density (and since the areas are equal, these blocks *would* be comparable). That’s a cartographic question, but the fact remains that until you remove the effect of the area, the absolute value is misleading and shouldn’t be used.

          4. Have to admit, I was not expecting my choice of mapping techniques to be the most controversial thing about this post!

            Hamish, I appreciate your expertise in the area, but I think you’re overengineering things. All the map is trying to communicate is that there is a large quantity of jobs outside of Auckland’s city centre. I could have presented the same data in a bar chart showing (for example) the number of employees in each Auckland local board.

          5. I disagree that spatial intensity of jobs which is what you are actually talking about is the message.

            Peter doesn’t want you to know that there are 100 jobs per acre in wiri. He wants you to know that there are 30000 jobs. You kbow that now. You would not know that if density were the choropleth variable

          6. The bar graph would be less misleading because, presumably, you wouldn’t in a bar graph use darker colours to indicate larger absolute numbers. Instead you would use only the larger bar size to communicate this. It’s a common complaint of colour use in bar graphs, if some colours are more intense than others, overemphasising those particular bars, and i think your original graphic falls into this trap in the same way, larger area and more intense colour, suggesting a greater difference than what exists. To me your graphic suggests there is massively more employment in the airport region than the CBD. Nice article nonetheless!

    2. I did wonder why the graphic made it look like waiheke was such an employment hub. But now i see, it’s just that it’s a large area. Likewise with that large block in the north west harbour around whenuapai.

  6. Nice analysis. But you have fallen foul of the “flaw of averages”: the average distance to a job is not the same as the distance to the average job. Some lucky Aucklander has the average job in their own living room (or very near it); that doesn’t mean they have no commute.

    1. If you read the report from which the first map (average commute distances for Auckland area units) was taken, you will see that it’s based on journey to work data from the Census, which measures people’s _actual_ commutes.

      While there is still some averaging going on, it’s happening on the neighbourhood or suburb level. Obviously, there will be some heterogeneity in commute distances within areas.

      For example, the average commute distance in Parnell is roughly 5km. That means that about half of Parnellians commute shorter distances – probably to Newmarket or the city centre – while half commute further than 5km. In some cases, like Parnell resident John Key, their commutes might be quite a bit longer than the average.

      I did, at one point, take a look at the distribution of commute distances within Auckland area units. While a full analysis would have been computationally unpleasant, the suburbs I examined in different parts of the city seemed to show a similar distributions travel distances – few people travelling very short distances, most people travelling distances close to the suburb-level average, and a long tail of long-distance commuters.

  7. I find the concept of an average job location a bit confusing. Would I pay anything extra to live close to the domain to be midway between possible future work locations? Probably not. I would pay to live in a good area, close to good schools and definitely top be close to a beach or have a view from a north facing slope. But future workplaces I might never want to go to? And an average tells you so little. It might be the furtherest point from any individual workplace if they are around the periphery. Sometimes an average can be totally meaningless sort of like betting on 19 or 20 in roulette because it is the average. You have concluded that being central is important because that is where the centroids are. You conclusion is actually part of your assumption so I think you are ‘begging the question’ in its logical meaning

Leave a Reply