Review - Human Transit: How Clearer Thinking about Public Transit can enrich our communities and our lives, by Jarrett Walker

An observation from last September’s Australasian Transport Research Forum conference was the abundance of academics versus the paucity of practitioners.

I define practitoners as those who plan and operate transport routes or closely work with those who do. They have their own work and don’t write books and attend conferences as much as academics and bureaucrats.

Consequently it was refreshing to find a book on public transport service planning by an experienced practitioner. The author, Jarrett Walker, is a service planner and consultant to transport agencies in the US, Canada, Australia and New Zealand. His work was already known to me through his blog at humantransit.org

Having read books by academics such as Mees, Newman and Kenworthy, I was particularly interested in the different perspective that a practitioner might bring.

Academics thrive on research, teaching and (sometimes) political advocacy. However the affordability, practicality and acceptance of their recommendations may not frequently be tested.

Whereas practitioners need to be more practical and less theoretical. They must often work within existing resources; something academics might find limiting. Consultants can be practitioners too but won’t get work if not seen as useful.

In Human Transit, Walker describes himself as a plumber.

We don’t expect plumbers to tell us how to run our cities and our lives; unlike (say) priests, politicians and talkback radio hosts. They may hold values but cannot choose them for their clients. But they might be able to advise on the consequences of particular decisions; for instance the cheap quick fix versus the dearer but enduring repair. And they might point out that any option chosen must respect basic laws, like water flowing from high to low, for the system to work effectively.

City planners and developers deal the hand with which transit planners must work. Some decisions, notably with respect to density, are controversial. Walker, more than Mees, sees density as conducive to higher patronage. However he adds, correctly in my view, that location and geometry are equally critical when designing neighbourhoods for direct and efficient transport routes. To Walker these factors are public transport’s equivalent of the plumber’s basic laws. If disobeyed an area may never be easily or efficiently served by good public transport.

Transport networks themselves can be optimised to favour patronage or coverage goals. Networks that optimise patronage generally have direct, fast, frequent but more widely spaced routes, whereas a coverage-oriented network sacrifices directness, speed and frequency for short walking distances. Walker insists that transport authorities set a resource allocation policy between ‘patronage’ and ‘coverage’ routes.

Only after such priorities, and thus available resources, are known can the service planner design the network.

The book is a mine of simple but often neglected insights that would make public transport better. All are applicable to Melbourne. For instance the trade-off between directness, frequency and legibility on the one hand, and requiring that people change for some trips. We’re used to road maps having different thicknesses for different types of road, from lane to freeway, so why not do the same with transit maps, highlighting our most frequent,and thus most usable routes? And, on busy roads, the contribution of the humble pedestrian crossing to improving access to bus stops and the impetus it provides to gradually increase an areas’s walkability (and thus transit access).

Human Transit does what its title suggests. It forces the reader, from layperson to transport planner, to sharpen their thoughts and ask the right questions. I highly recommend it.

Human Transit is published by Island Press and is available via the link below.

The user time element in public transport service planning (Pt 2)

Can it be done?

Still think that substantial travel time reductions are impossible without building high-speed railways under each suburb?

The following examples show how significant travel time savings could be possible, largely with existing infrastructure.

Example 1: Inter-suburban bus trip

Consider a short bus trip where the bus runs every 40 minutes. This service level is typical for middle and outer suburban routes and even some major inner suburban routes (eg 246 on Sunday evenings).

The trip takes ten minutes, with five minutes allowed to reach and leave the stop at either end. That adds up to a best case 20 minutes travel time, if you arrive at the stop just as the bus arrives. The worst case, if a bus has just been missed, is 60 minutes. While the average, assuming random arrival is 40 minutes.

This comparison shows high variability, or +/- 20 minutes from the average. The 3:1 difference between maximum and minimum trip times is entirely due to the attempt to make a short trip by randomly arriving to catch a low frequency service.

Increasing frequency to 20 minutes reduces the average time to 30 minutes, or a 33% reduction. Variability is reduced to +/- 10 minutes, or 20 to 40 minutes. For a ten minute frequency the average drops to 25 minutes, or an average 37% time saving. Variability drops to +/- 5, or a ratio of maximum to minimum of 1.4:1.

The above example shows that frequency has a dramatic effect on random arrival end-to-end travel times. Higher frequency also cuts variability, or effectively increasing travel reliability. The minority willing to plan around timetables can enjoy the benefits today, without any frequency increase. That’s unless they depend on connections, which are discussed next.

Example 2: Bus + train trip

Consider a suburb about 20km from the CBD. Its local buses run every 30 minutes and the trains every 20 minutes. The passenger is 5 minutes walk away from the bus stop. The bus trip (via a meandering route) takes 15 minutes to a stop near the station. When passengers alight the bus they need to cross a busy road to the station, which has only one entrance at the far end of the platform (assume 5 min). After that the train takes 40 minutes to the city.

The best case travel time is 5 min (walk) + 0 min (wait for bus) + 15 min (bus travel) + 3 min (station access) + 0 min (wait for train) + 40 min (train travel). Or a total of 65 minutes for the 20km trip.

The worst case travel time is 5 min (walk) + 30 min (wait for bus) + 15 min (bus travel) + 5 min (station access – assuming 2 min traffic light cycle at crossing) + 20 min (wait for train) + 40 min (train travel). That’s a total of 115 minutes for the 20km trip.

There’s three things to note. Firstly the worst case represents an overall speed of just over 10km/h or about double walking speed. Secondly the variability is high – the worst case taking twice as long as the best case. And, in the worst case example, the passenger is in motion for barely half the time.

Now consider the same trip above, with the following modest service improvements:

· Bus frequency upgraded from 30 to 20 minutes, to provide a harmonised connection to each train, with a consistent 6 minute connection · Bus route made more direct, to reduce travel time from 15 to 10 minutes and fund the higher frequency, but with 5 minutes added walking time · Additional station platform entrance and zebra crossing installed (to reduce station access time from 5 to 2 minutes)

The best case travel time following these improvements is as follows: 10 min (walk) + 0 min (wait for bus) + (10 min bus travel) + 2 min (station access) + 4 min (wait for train) + 40 min (train travel). Or a total of 66 minutes for the 20km trip. The worst case following these improvements is as follows: 10 min (walk) + 20 min (wait for bus) + 10 min (bus travel) + 2 min (station access) + 4 min (wait for train) + 40 min (train travel). Or a total of 86 minutes for the 20km trip.

The difference is dramatic. The average travel time has fallen from 90 to 76 minutes, while the ‘worst case’ is nearly 30 minutes quicker. Variability also fell; from +/- 25 minutes of the mean to +/- 10 minutes. Better connectivity and higher bus frequency contributed most to the gain. However more direct bus routing and better pedestrian access also added smaller but no less cost-effective benefits.

Example 3: Bus + Bus trip

Finally we’ll examine a cross-suburban trip involving a change between two bus routes. This is typical for journey types in which public transport has a low modal share.

I’ll use similar assumptions to the first example. Eg 5 minute walk to and from the bus and 10 minute travel time in each bus. The first route runs every 60 minutes while the one being changed to is every 40 minutes.

The first leg involves 5 min (walk time) + 30 min (average wait) + 10 min (travel time), or a total of 45 minutes. The best case is 15 minutes, while worst case is 75 minutes. Or a variability of +/- 30 minutes.

The wait to the second bus will be anywhere between 0 and 40 minutes. Because the frequencies are unharmonised the best connections will recur every two hours. We’ll assume an average of half its frequency, or 20 minutes.

The second leg involves 20 min (average wait) + 10 min (travel time) + 5 min walk time, or a total of 35 minutes average. But it could range from 15 to 55 minutes, or a variability of +/- 20 min from the average.

The very shortest time that the overall trip can be made is 30 minutes, with the longest 130 minutes. The average time is 80 minutes. Because time savings and delays average out, the traveller is unlikely to experience the extreme shortest and longest trip times. But if they do, that’s a ratio of over 4:1, or a variation of +/- 50 minutes.

While people may tolerate a higher variability for a short trip (Eg a 10 minute trip taking 20 minutes), it is probably true that tolerance declines for longer trips (especially if routine). Hence the letters in the paper complaining about suburban trips that take an hour by public transport but only 20 minutes driving.

There’s a couple of things that can be done to reduce variability.

Firstly the passenger could forego flexibility and use a timetable. Instead of waiting an average of 30 minutes for the first bus, they wait an average 5 minutes. This reduced variability of +/- 20 minutes is solely due to the connection between the first and second bus, which is beyond the passenger’s control. Average travel time is also reduced – by 25 minutes, which is the difference between the planned wait and the random arrival wait (ie half the frequency of the first service).

There’s also the contribution of service planning, which unlike the first response, assumes no accommodation on the part of the passenger.

Suppose the frequency of the first service was upgraded from 60 to 40 minutes. The first benefit of this is to reduce the average wait, from 30 to 20 minutes. Variability contributed by the wait for the first service is thus reduced.

The second benefit is that it matches the frequency of the second service. Such matching does not guarantee good connections but does dramatically slash variability. Let’s look at the numbers.

The first leg involves 5 min (walk time) + 20 min (average wait) + 10 min (travel time), or a total of 35 minutes. The best case is 15 minutes, while worst case is 55 minutes. Or a variability of +/- 20 minutes.

The wait to the second bus will be anywhere between 0 and 40 minutes. Because the frequencies are now harmonised the connections will recur every 40 minutes. We’ll assume there’s been no special planning and the wait for the second service is half its frequency, or 20 minutes.

The second leg therefore involves 20 min (average wait) + 10 min (travel time) + 5 min walk time, or a total of 35 minutes. Because the first leg has been harmonised to it the wait for it is now constant, variability has been reduced to zero.

Add the two legs and we have an average of 70 minutes. That’s 10 minutes down on the first case of 80 minutes. But the real gain has been in reduced variability. At best it’s 50 minutes and at worst it takes 90 minutes. This is a variability of +/- 20 minutes – well down on the earlier +/- 50 minutes. Also the ratio of maximum to minimum journey time has fallen from over 4:1 to under 2:1. Although average travel times are still slower than many would like, the improvement made from adjusting one route from a non-harmonised 60 minutes to a harmonised 40 minutes cannot be underestimated.

Again, with the earlier example one can do better. If one sacrifices flexibility and uses a timetable to catch the first service, the average travel time falls by 15 minutes (70 to 55 minutes) and variability virtually eliminated. Secondly, if planners consider that the connection between the two services is sufficiently important to be worth adjusting timetables, the connection time could be reduced from the 20 minutes average assumed here to 10 minutes. This contributes another 10 minutes, meaning a total average trip time of 60 minutes for those who don’t use a timetable and a reliable 45 minutes for those who do.

Summary and Conclusion

I have demonstrated the effect of frequency on cutting journey time. It is at first dramatic, with a point of diminishing returns being reached as frequency rises to around ten minutes. Beyond that point, unless it needed for capacity or for very short trips, its impact drops.

Also discussed has been travel time variability. Public debate on this normally concerns train reliability, and this is especially important for those connecting to less frequent buses. However the examples demonstrate indicate it can be very high for bus trips, especially those involving random arrival and connections between non-harmonised services. Harmonised bus frequencies can greatly reduce variability and make public transport more useful for trips where it’s currently weakest.

The planning approach presented here focuses most on service frequency and its harmonisation. There is less attention to infrastructure and capacity.

The former is cheap and quick, while the latter is expensive and long-term.

Both have their place in a growing city. But introducing the latter without the former means that use of the latter is poorly utilised and public transport’s potential to fully contribute to the overall transport effort is unrealised.

The user time element in public transport service planning (Pt 1)

There’s been heavy discussion about the financial user cost of public transport, following the announcement of an 8.6 per cent fare rise from next month. This post is about another user cost; the longer time it sometimes takes relative to driving.

Amongst those with a choice, public transport attracts its highest share where it is time competitive with driving, notably work trips to the city. Where it’s uncompetitive passengers tend to be those with more time than money, typically due to low incomes and/or an inability to pay. This is the familiar skewed pattern of old and young in the off-peaks and city commuters during the peaks.

This is a long post, so is split into two parts. Part 1 discusses user financial costs, funding sources, trip time and trip time variability. Part 2 uses examples to illustrate the large gains possible from service and other improvements.

Two dimensions: time and money

One insight that comes from counting time as a cost is that it invites comparison with that other type of cost most commonly associated with public transport – fares. Fares are known and fixed whereas time value is less tangible but no less important.

Although lacking hard numbers, we can guess time value’s magnitude or draw inferences from surveys and modal share statistics. We know it varies with trip length, location and timing. Acceptance of slow trips and high costs varies but depends on the value placed on time and money respectively. Some rough plots of user time versus financial costs for various transport modes in Melbourne are on the graph below.

The ‘best’ is cheap and fast (bottom left) while the worst is dear and slow (top right). The other corners are occupied by fast expensive (top left) and slow cheap (bottom right). Overall user costs for each comprises dollar plus time costs, with this increasing from bottom left to top right.

Consensus that a particular mode represents good value is highest for those near the bottom left. Whereas top right is poor value and only worthwhile for those willing to bear the high overall cost. If they are not, they’ll change homes, jobs or schools to avoid it. The action is crucial; mere grumbling implies acceptance since bad transport may still be better than other choices available.

I take the view that most people are transport pragmatists, ie whichever available mode suits their time / cost value profile for a particular trip will be used. Most of the time the mode chosen (driving) is faster but dearer than public transport. This invites the question as to whether the overall user value of public transport would increase if it were faster, even if somewhat dearer.

Public transport’s place

Public transport’s detractors treat slowness as inherent but this is not necessarily so. Where transit is slow, this is due to decisions made on station spacing, route and timetable planning, street design, signal priority and allocation of road space between modes.

Skybus (also on the graph) is an example of fast public transport. Its fares are high relative to government-subsidised routes. But because it’s both fast and frequent its time costs are low. Add the two and you see that Skybus occupies a distinctive (and successful) position in the market, set apart from regular buses (with low fares but high time costs).

Is regular public transport’s position on the graph optimum?

The answer depends on what you want it to achieve.

What would happen if you changed either its user financial or time costs?

Making an already low-priced but infrequent bus service cheaper would not greatly reduce the overall user cost to the ‘average person’. This is because unless they value their time lowly, the time cost component probably outweighs the fare cost and slashing the latter won’t cut the total much.

However this conclusion ignores certain market segments. Fare cuts are more significant for those who value money more than time, ie the ‘captive passsenger’ end of the market. This segment tolerates the limited service and values the saving more than average so will probably use the service more, possibly even increasing revenue.

It’s a bit like arguments in favour of Reaganomics tax cuts; individuals pay less but incentive stimulates economic activity which increases overall tax take. Fare / patronage elasticity will likely be highest off-peak and weekends as there’s more discretionary trips made by a different passenger profile.

The above limited bus service isn’t attractive to car owners unless driving conditions are very poor. People on all but the lowest incomes may find the waiting not worth the saving. Passengers that a lowered fare attracts may have previously walked rather than driven. Cheaper buses might make individual travel more energy intensive, although public transport fuel efficiency (as measured by fuel use per passenger) improves with patronage (ie more passengers per litre).

Speeding up buses (achieved through directness, frequency and connectivity) but charging a higher fare should have a different effect. The better service may attract some who previously drove or got lifts. Existing passengers may use the improved, more capable service more. But a fare increase could mean that even though the service is better, price-sensitive passengers abandon it for their lower value discretionary trips.

Funding better transport

There’s at least three things that can be done.

* The first is to maximise effective service levels from current budgets. For example delete routes that duplicate others and use saved resources where more worthwhile. This requires a strong service planning culture, and for co-ordinating agencies to take a network view; both factors in which Melbourne has been weak. Like tidying an over-grown garden before selling your house, good service planning has high payoffs, due to this legacy. And participatory public engagement, including illustrating the service gains possible, should help build acceptance.

* Secondly there’s the possibility of differential off-peak pricing. This gives the price-sensitive a discount while preserving revenue from the less price sensitive. This segments the market and probably optimises patronage, though at expense to legibility. A difficulty here is if peak fares must rise disproportionately to retain overall revenue, and here we merge into the next point.

* The third are other measures to generate revenue. Possibilities include development of rail air space, parking taxes, hypothecated property levies or above-CPI fare increases.

All are politically difficult, especially if the fare increase has not followed tangible reliability and service improvements. Rises skewed toward peak period commuters (without whom capacity building infrastructure would not be needed) may be economically most rational, but send the wrong signals with regards modal choice if not accompanied by similar increases for driving (whether through fuel, parking or road space pricing).

This combined approach may be better transport policy but probably harder than fare hikes alone due to the larger numbers affected. While not impossible, it may require a trust in government that is lacking, particularly in states that propose then cancel major transport projects, such as New South Wales with its various metro plans.

Whereas people are willing to individually pay more for quicker travel (as witnessed by the popularity of driving versus public transport), sourcing such funds for collectively funded network improvements raises hackles. How do we break this deadlock?

Seeing improvements differently

It may be worth reviewing how we regard transport improvements.

Large transport plans may be centred on infrastructure. Its purpose is typically to relieve ‘bottlenecks’ caused by extrapolating present commuting patterns forward. Time costs may be used to justify it; studies may cite rising congestion costs and ‘lost productivity’. Infrastructure’s tangible nature invites us to draw lines on maps and build models to imagine a faster future.

Another way of thinking is to set a public transport journey time reduction target and use the best combination of infrastructure and services to achieve it. The bigger the target the higher the cost, but the more likely that public transport would win the modal share shifts envisaged in Melbourne 2030 (an increase from 9 to 20% or a patronage trebling).

A 20 to 50% cut in average public transport travel times may seem far-fetched. But considering how it could happen would sharpen our thinking about which improvements are really worthwhile. Plus it familiarises planners with the real source of delays, which passengers know but they may not.

Overall travel time and its variability

Rules for such analysis need to match where and when people live and travel. For example they must acknowledge that people live where they do and not at railway stations. They want to leave when they want, without checking a timetable. And the selection of trips used must represent travel made by all motorised modes, and not skewed towards public transport’s current profile. These rules are observed by (i) counting waiting and transfer times, (ii) assuming random arrival at the stop or station, and (iii) including a representative spread of trips (including cross-suburban, night and weekend travel, which together form a majority of all trips).

Adding the first two produces a more useful average travel time (random arrival end to end travel time) than ‘headline’ minimum in-vehicle travel times. The latter is often a dubious selling point for ‘fast rail’ projects. It’s a bit like quoting air fares without all the extras and surcharges; meaningless at best and misleading at worst.

Travel time variability also needs to be known. Which transit system is better? One where a particular trip takes anywhere between 10 and 40 minutes. Or one that at best is slower (eg 15 minutes) but at worst is faster (25 minutes). It’s no contest really; the latter’s lower variability means better reliability, especially for time-critical trips.

Average random arrival end-to-end travel time and travel time variability (for a representative sample of trips in the area) are the to key numbers planners need to pick the best from a number of route and timetable options.

Train planners may have to juggle with express running versus frequency. Service delivery is also critical; when passengers start having to catch an earlier train due to the risk of the one they want being cancelled, effective travel speeds can easily halve, especially when a connection from a bus is involved.

Tram planners have little scope for express running and frequency is generally already high. Their main problems are externally caused, especially when in mixed traffic. Although there are trade-offs between whether all services are run to the terminus or some terminate early to provide greater frequency and capacity in the busier inner core.

Bus planners have more flexibility. They must balance walking time, coverage, number of routes, use of transfers, directness and frequency. However, like trams, priority on the roads can speed travel and reduce variability. This is especially where bus priority operates at all times and not just when the bus is running late.

Part 2 will follow with examples

The end of Melbourne's least legible bus route

Melbourne bus passengers will get less lost after the cessation of Melbourne's least legible bus route after end of service today.

Route 672, in Melbourne's outer east, sometimes runs between Croydon Station, Wonga Park and Chirnside Park Shopping Centre. I say 'sometimes' because at certain times of the day almost each trip goes a different way.

672's proliferation of timetabled deviations meant that buses often missed stops, finished at unpredictable locations, went down other streets and even went on the opposite side of the road to what maps showed.

As attested by the number of footnotes, Route 672 had no standard pattern. Footnotes appear on timetables at stops. But if you're not up for a trip to Croydon, you'll be able to read them online for the next few days.

Current 672 timetable  (subsequent archived link)

The first thing to read is the route description. Note how the variations take up more space than the route route itself. If you're game, try to follow these on the map. Finally inspect the timetables themselves (either the pdfs under download route timetables or individual stop timetables). One direction has no less than nine footnotes and the other no less than six.

Compare it with Monday's timetable below.

672 timetable from 5 December 2011 (subsequent archived link) 

The description, map and timetable for the revised 672 are vastly simpler. Except for the Telebus deviate-on-demand service which remains, fine print footnotes have been almost eliminated. There are later finishes on weeknights and many more Saturday trips.

Other changes

The 672 isn't the only route that will be simplified on December 5. A good summary appears here.

Key cases where occasional deviations and extensions were removed include 691's Bayswater and Monash extensions (the latter now covered by other routes) and 664's Sunday deviation to Lilydale (a hangover from when 664 and 691 were about the only Sunday bus routes operating east of Stud Road). Other routes eg 671, 680 and 693 also lose occasional deviations or are straightened.

Span and frequency

As well as legibilty gains, there have also been span and frequency improvements.

The most common span improvements include one or two later weeknight trips and a fuller Saturday afternoon service. Frequencies have been made more regular on several routes. The largest changes include (i) a doubling of Route 693's off-peak service to 30 minutes, (ii) 664 becoming a 30 minute service along its full length (it was previously a split 20 / 40 minute service, unharmonised with trains every 30) and (iii) some weeknight frequency improvements between the pm peak period and last service around 9pm.

Overall the changes in the Invicta network reflect the biggest improvement since some routes in the area got minimum standards upgrades. They deliver substantial legibility gains without greatly changing the existing network. It's an example of a 'greater good' upgrade that should result in a net overall benefit.