Does density matter for mode share?

Mode share vs density (click to enlarge)

The accompanying chart shows how public transport’s share of the journey to work varies with population density across 41 US and Australian cities.

It is taken from the same article that I mentioned in my last post. The authors, Dr John Stone and Dr Paul Mees, find there is only a modest relationship between population density and transit share (R2 = 0.229). They conclude that “higher density across the whole urban region is not the explanatory variable that many might expect”.

Los Angeles, for example, is the densest metropolitan area in the US – denser ever than New York – yet the chart shows public transport’s share of work travel in LA is much smaller than in NY.

If that seems counter-intuitive, your intuition could be right. The chart uses average population density calculated across the entire urbanised area of each city.

While that’s perfectly alright in some contexts, it doesn’t allow for the possibility that public transport’s ability to win travel away from cars is related to the morphology of density – the ‘peaks and troughs’ in the way the population is spatially distributed. It’s possible that the relative proportion of population in high density areas vs low density areas has a greater impact on mode share.

Using average density probably won’t present a serious problem with cities like Atlanta, Austin, Dallas, Phoenix and Portland where the population is overwhelmingly suburbanised at relatively uniform (low) densities. But it could have a big impact on places like New York which have an extensive ring of low density suburbs as well as a high density central region e.g. Manhattan and Brooklyn.

A way of dealing with this issue is to use weighted density rather than average density. This involves weighting the density of each suburb (or other convenient geographical unit e.g. traffic zone) by its share of the city’s total population. So a one km2 suburb with 5,000 residents (say) carries a lot more weight than another suburb of the same area that has only 1,000 residents.

Weighted vs average density (2000)

Fortunately, both weighted and average densities have been calculated for a number of major US cities by Texas resident, Chris Bradford. The second chart, which is based on his data, shows that some cities rank much higher on weighted density than on average density.

Los Angeles slips from 1st place on average density to 3rd place on weighted density and Denver, Phoenix, Houston, Riverside and Portland each move down several places in the rankings. Conversely, New York jumps from 4th to 1st place and Chicago, Philadelphia and Boston – all cities with relatively dense cores – each move up multiple places.

It is also notable that the ‘distance’ between cities changes dramatically. Using average density, the top ranked city is only 1% denser than the second ranked city and four times as dense as the bottom ranked city. However with weighted density, New York is 65% denser than second ranked San Francisco and eight times as dense as sprawling Atlanta.

The mode share data in the first chart seems to gel better with our perceptions when weighted density is used (although caution is needed in comparing the two charts because they are based on different years and the definition of city boundaries might not be the same). But even accepting there’s a stronger relationship between these two variables than the first chart suggests, there’re other possible drivers of mode share.

These include factors like income, but one of particular interest is the density of non-population activities. Since we’re discussing the journey to work, the relevant measure is employment density.

The five cities in the US with the highest (job) weighted employment density are New York City, Chicago, San Francisco, Boston and Washington DC. These are also the US cities which have the highest mode share in the first chart and are towards the top of the weighted density rankings in the second chart.

What seems to matter most is the concentration of employment. Consider the case of Melbourne, which the first charts shows has relatively high mode share in this company. The densest 10% of jobs in Melbourne occupy just 0.1% of the built-up area. The least dense 10%, on the other hand, occupy 51%. Public transport wins around 70% of work trips to the CBD, but just 15% across Melbourne as a whole. (I define the built-up area as traffic zones with density >50 jobs/km2. This gives an area just over a quarter the size of the Melbourne Statistical Division).

Thus at the metropolitan scale, there seems to be a strong relationship between (weighted) density and mode share in US and Australian cities, at least for the journey to work.  The density of both population and employment matter, the latter probably more so.

In a city like Melbourne where the dominant sentiment is pretty well to leave suburban residential densities alone, high employment density can have a real impact on mode share. Obviously that depends on adequate public transport being provided, but it also depends on the density and scale of job centres being high enough to create the congestion that makes cars less competitive. And there’s another whole issue about how many and what sort of firms actually want to locate in high density centres.


8 Comments on “Does density matter for mode share?”

  1. […] Posted on February 6, 2011 by hrgh| Leave a comment Over at The Melbourne Urbanist, Alan Davies picks apart some of the arguments from this report, and takes the analysis […]

  2. rohan storey says:

    Thanks for that weighted density stuff – shows why the first graph didnt make a lot of sense if you actually know what those cities are like. Now just need someone to make new graph …. love to see where melb comes out !

  3. jack horner says:

    Remember Newman/Kenworthy’s graph of population density against gasoline use per person in different cities? The result is hyperbola – high density, low gasoline use on one arm, & vice versa on the other. See http://www.patrec.org/web_docs/atrf/papers/1994/Brindle%20(1994).pdf

    At the bendiest point of the hyperbola (about 30 people per hectare) the reduction in gasoline use per increase in density appeared greatest. So people jumped on the idea that this was a tipping point – get density up to this and great things would start happening.

    Looking at the graph again 20 years later, I see something quite different: on the vertical axis many cities of similar density have vastly different per person gasoline use. Why this is so might be the more interesting story.

    PS I’d be grateful for more explanation of exactly how ‘weighted density’, the middle column on the graph, is calculated.

    • Alan Davies says:

      What an interesting paper! Thanks for the link. Here’s the revised version of the graph that was done (presumably by Newman and Kenworthy) to address criticisms of the original. The Wiki discussion is worth a read.

      You can see an explanation at Chris Bradford‘s site of how weighted density is calculated. Otherwise, it’s:

      {the sum of (the population density of each zone x the absolute population of the zone)} divided by {the sum of the absolute population of each zone}.

  4. Tim says:

    Hi,

    You should read Mees’s book you realize that the reasoning for his argument is to attack the ‘density as destiny’ rational that leads to a unwillingness to provide more than a token PT services in low density suburbs because ‘they are not dense enough’.

    The argument of average versus weighted density is a valid one but I think it does not undermine the guts of his argument. You would need to factor in the employment densities for Australian, Canadian and European cities to try and take into account the different Public Transport operational regimes.

    If you you don’t follow Jarrett Walker’s blog, he also discusses the average versus weighted density argument in the posts can we make density make sense? and the perils of average density.

    • Alan Davies says:

      I will read it….but there’s only so much time and there’s so much to read. I rely a bit on journal articles – the Stone and Mees piece has only just been published so it should be in line with their general views.

      I have consistently argued that density should not be used as a reason for maintaining sub standard PT services in the suburbs. But it’s still important to understand how important density is for PT at the metro scale.

  5. jack horner says:

    Between wars, vast areas of what are now middle ring suburbs in Sydney and Melbourne were developed at densities little different from new urban fringe suburbs today, with low car ownership (obviously, considering the time) and high public transport use (obviously, as a result). So clearly it is physically possible to work a city that way, should one wish to do so. The challenge is to recreate that type of city in a world of high car ownership.

  6. […] 11 of land use and 15 other, mainly demographic, factors. Moreover, he employed the concept of ‘weighted density’ (he calls it ‘perceived density’) to more accurately describe the distribution of both […]


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s