MRCagney Works
Job Accessibility
Recently the Data Science Team conducted a study of multimodal job accessibility in Hamilton, New Zealand, the purpose of which was to test our tools on a timely topic of interest to our clients. Let me walk you through our approach and findings.
We chose Hamilton, New Zealand as our study area, because it is populous enough to be instructive and small enough to compute everything quickly on our laptops. By Hamilton, we mean the area within the Hamilton City Territorial Authority as shown below.
We followed the sensible approach of Conveyal and measured job accessibility with travel time isochrones. In doing so, we aimed to answer, at least approximately, the following questions for a given mode of travel, e.g. public transport, and a given time bound, e.g. 30 minutes.
- Given a point in Hamilton, how many jobs can you reach from that point by that mode and within that time bound?
- What percentage of Hamiltonians can reach what percentage of jobs by that mode and within that time bound?
The first question addresses the geographic distribution of job accessibility in Hamilton and the second the statistical distribution.
Question 1
There are infinitely many geographic points in Hamilton, which is too many to compute isochrones for, so we whittled our source points down to a manageable finite set. Using all street address points in Hamilton was one option, but there are 64,859 of those, which is still too many for our laptops. So instead we partitioned Hamilton into a somewhat fine regular grid, chose one sample point from each grid cell, and used that point’s accessibility score to represent the accessibility score of the entire cell, a reasonable approximation.
We chose a hexagon grid with inradius 250m (the radius of the inscribed circle of each hexagon), because we like honeycomb and reckoned that those cells would be small enough to yield meaningful results but few enough (513 of them) to yield quick computations. For the source points we chose the most central address in each cell.
Here is what that grid looks like, clipping it to Hamilton’s boundary and removing cells lacking residential addresses. To see the source points, toggle them via the map’s layer control in the top right.
Gather job locations
Now where are the jobs? The finest grained open data we knew of to answer that question is the statistical area 1 data set for 2018 Census. That data set divides Hamilton into statistical area 1 cells (SA1s) and groups job locations by them. Here is a choropleth map of the number of jobs by SA1.
What I am calling the number of jobs is more precisely the number of jobs filled by workers aged 15 years and up and is the Census_2018_Industry_by_workplace_address_Total_CURP_employed_15years_and_over field of the census data.
But how many jobs lie in our hexagon cells? Those are the cells we are working with after all. To answer that, we overlaid each SA1 onto our hexagon grid and distributed its job count across the hexagon pieces it covers in proportion to the fraction of the SA1 area the hexagon pieces take up. That is called resampling by area-weighted average and is a standard geospatial move. It will not give us completely accurate hexagon job counts, because it assumes that the SA1 jobs are spatially evenly distributed throughout each SA1, which is false. But it probably is not too false. Finer source data would give us more accurate hexagon job counts, but that is the finest source data we could find.
Here is a choropleth map of job counts resampled to our hexagon grid.
Compute isochrones
Next we computed isochrones, one for each of our source points and each travel mode among walking, cycling, driving, and public transport. For the time bound on our isochrones we chose 30 minutes. That is a good upper bound, because most Hamiltonians spend less than 30 minutes commuting to work one way. Indeed, New Zealanders spend about one hour per day traveling for all purposes combined (commuting, shopping, recreating, etc.). Moreover, this one hour per day seems consistent across the world and over the last several thousand years! See Marchetti’s constant.
Here are four filled-in isochrones for one of our source points, one for each travel mode. Here “transit” abbreviates “public transport”.
A few more details on these isochrones: the driving ones use free-flow traffic conditions and the transit ones use an 08:00 departure.
Count jobs in isochrones
Next, for each source point, each travel mode, and each 30-minute isochrone starting from that point and using that mode, we counted the number of jobs contained in that isochrone by resampling by area-weighted average the hexagon grid job counts into the isochrone. Then we attributed the job count for each isochrone to its source cell as an accessibility score for that cell.
Here are choropleth maps of the accessibility scores for each mode of travel, answering Question 1.
Question 2
On to Question 2: What percentage of Hamiltonians can reach what percentage of jobs by that mode and within that time bound? In other words, how many people, more specifically workers aged at least 15 years, experience the levels of job accessibility mapped in the choropleths above?
With our set up, we could answer that question, at least approximately, fairly easily, by counting how many workers reside in each hexagon cell and attributing the cell’s accessibility score to all those workers. Then we segmented the accessibility scores into deciles and counted how many workers experience those job accessibility score ranges.
Here are bar charts representing just that.
As you can see,
- 30-minute walking access to jobs is relatively low with 85% of the working population having access to at most 10% of jobs
- 30-minute cycling access to jobs is middling with about 54% of the working population having access to at least 50% of jobs
- 30-minute driving access to jobs is amazing with 100% of the working population having access to 90%–100% of jobs
- 30-minute transit access is worse than cycling access in that only about 5% of the working population has access to at least 50% of jobs.
Finally, when a single number summary of each mode’s job accessibily is preferred, we computed the population-weighted median, which is less susceptible to outliers than the more popular population-weighted mean. For Hamilton, these medians by mode are shown in the following table.
| travel mode | median job access (%) |
|---|---|
| walking | 3.7 |
| cycling | 52.9 |
| driving | 100.0 |
| transit | 28.8 |
For example, we see from the table that half of the working population of Hamilton can access at least ~29% of jobs within a 30-minute public transport ride and half can access at most that percentage of jobs.
Using this kind of analysis, particularly when comparing future scenarios, policy makers can make and assess accessibility targets to better choose how to invest in their city’s transport infrastructure.