A few months ago, I wrote about my search for all of the new money and transit service our beneficent leader, John Tory, had showered on Toronto’s transit riders. The answer, of course, is that the amount delivered is far from what was promised, and that’s before the pending cutbacks of the 2017 budget.

Recently, I was looking at the TTC’s 2016 Customer Charter in preparation for an article about the Five Year Plan (coming soon), and I noticed that a few promises and claims don’t line up with what is actually happening. During a period when TTC ridership growth is minimal and planned service improvements have been deferred, there is a surprising Third Quarter achievement:

We will add service during peak periods to 25 busy bus routes, to reduce crowding and improve travel time.

This prompted some head scratching considering that recent service changes have all been about trimming service, not making major additions, and certainly not to 25 busy routes. After a few emails to and fro with the TTC, there was finally an admission that, oops, those new buses really don’t exist. Or, well, they did, briefly, but they don’t any more.

In January & February 2016, the TTC increased peak bus service on 28 routes using 23 buses thereby achieving the 2016 charter commitment early.

However, since then, an equivalent number of peak service reductions have been made to mitigate the impact of the TTC’s deteriorating and decreasing legacy streetcar fleet and to sustain service in the downtown. In November 2016, the TTC reduced service on 8 routes freeing up 9 buses to reallocate to the 511 Bathurst. And, in January 2017, the TTC reduced service on 13 routes freeing up 20 buses to reallocate to 501 Queen and other streetcar routes.

When we “put the tick” in the box saying we had achieved the improvements we didn’t know how bad things would get with respect to vehicle availability.  We’ll reflect the fact that we had to unwind peak service on some of the same service improvements when we publish our year end charter review. [Email from Deputy CEO Chris Upfold, Jan. 18, 2017]

There’s a bit of creative rewriting of history there, plausible though the explanation might be. First off, it was clear that when the 2016 charter was framed, the TTC knew that it had crowding problems, and planned to address them. Provision for this was in the fall 2016 service plan, but that was killed off when ridership failed to grow as expected through 2016. The improvements cited by Upfold, however, went into operation much earlier.

January & February 2016 – Peak period service increases (+23 AM peak buses)

• 191 Highway 27 Rocket – peaks and midday
• 63 Ossington
• 75 Sherbourne – both peaks
• 6 Bay – AM peak
• 21 Brimley – AM peak on 21A (Kennedy Stn-Scarborough Centre Stn)
• 20 Cliffside – AM peak
• 25 Don Mills – PM peak
• 143 Downtown/Beach Express – AM peak
• 144 Downtown/Don Valley Express – PM peak
• 125 Drewry – AM peak
• 111 East Mall – both peaks
• 195 Jane Rocket – AM Peak
• 83 Jones – PM peak
• 12 Kingston Rd – PM peak
• 44 Kipling South – both peaks
• 102 Markham Rd – AM peak
• 46 Martin Grove – both peaks
• 16 McCowan – both peaks
• 57 Midland – PM peak
• 65 Parliament – AM peak and midday
• 66 Prince Edward – both peaks
• 134 Progress – PM peak
• 123 Shorncliffe – both peaks
• 55 Warren Park – both peaks
• 94 Wellesley – AM peak
• 112 West Mall – AM peak
• 89 Weston – PM peak
• 11 Bayview – AM and PM peak

Service reductions came on a different list of routes, with some overlaps to the one above. The net effect was a reallocation of service rather than an overall improvement, although it could be argued that routes losing service had capacity to spare under the Service Standards.

November 2016 – Peak period service reductions to mitigate streetcar fleet constraints (-9 AM peak buses)

• 121 Fort York-Esplanade – AM peak
• 43 Kennedy – PM peak
• 129 McCowan North – AM peak
• 57 Midland – AM peak
• 133 Neilson – AM peak
• 124 Sunnybrook – PM Peak
• 168 Symington – AM peak
• 68 Warden – AM peak

January 2017 – Peak period service reductions to mitigate streetcar fleet constraints (-20 AM peak buses)

• 6 Bay – AM peak
• 14 Glencairn – AM peak
• 16 McCowan – AM peak
• 32 Eglinton West – AM peak
• 38 Highland Creek – AM peak
• 46 Martin Grove – AM peak
• 51 Leslie – AM and PM peak
• 85 Sheppard East – AM peak
• 102 Markham Rd – AM peak
• 112 West Mall – AM peak
• 123 Shorncliffe – AM peak
• 129 McCowan North – AM peak
• 190 Scarborough Centre Rocket – AM peak

The situation with delays in streetcar deliveries is a serious one that has been gradually building as the old fleet wears out. However, it is important to take a longer view of the evolution of bus service in Toronto to see how fleet usage has evolved over the years. The charts below, with data taken from the TTC Scheduled Service Summaries, show the total number of buses used during the two peak periods.

• The blue portion is for buses running on bus routes, and in recent years the green portion is the articulated bus count adjusted by 1.5 for the larger capacity of these vehicles. The top of the blue section represents the buses serving bus routes on an equivalent-to-standard-bus basis.
• The red sections at the top of each bar are for buses used on streetcar routes.

The AM peak service improved slightly in late 2015 thanks to new buses purchased that year, but the numbers have been falling back.The total AM peak has not been flat since mid-2016, but has been slowly falling. If this were only a case of trading off buses for streetcars, the total would be constant. This shows the effect of service trimming in late 2016.

The situation in the PM peak is different because this period uses fewer vehicles (school and work peaks do not coincide in the PM), and so service is not constrained by the fleet as in the AM.

As an historic note, the big jump in 2009 was thanks to the Ridership Growth Strategy and the increase in the peak fleet. Even in the Ford years, the TTC managed some service improvements mainly due to years in which a “flat lined” subsidy actually gave the TTC more money because they had run a “surplus” in previous years. (The “next year” budgeted subsidy was higher than the “previous year” actual subsidy draw.)

The number of buses now running on streetcar routes (AM peak) in January 2017 is:

• 501 Queen: 23 (Queensway/Humber rehabilitation)
• 502/503 Downtowner/Kingston Rd: 16 (Streetcar shortage)
• 504 King: 12 (Streetcar shortage)
• 511 Bathurst: 17 (Streetcar shortage)

Assuming that the 40 new Flexitys promised by Bombardier arrive during 2017, then in theory many of the buses could be freed up for service on their own network. However, the TTC might decide that retiring old streetcars is a more important goal and spread the return of full streetcar operation into 2018. Nothing is certain until we actually see Bombardier’s cars.

For budget planning, however, the City will have to face new costs in 2018:

• Full year effect of the Spadina Subway extension (estimated at $23 million net of new revenue)
• Restoration of bus service plus any new express routes and a Ridership Growth Strategy
• First year cost of the “Fair Pass” subsidy for low-income riders (estimated at $4 million)

Full streetcar service on 504 King (plus complete conversion of 514 Cherry to Flexity cars) will improve capacity on King Street, but much of this is unlikely until 2018 at best. Current plans for Flexity implementation are on 509 Harbourfront, 514 Cherry, 505 Dundas and 511 Bathurst.

The irony in 2017 is that the streetcar shortage is actually saving the city money because the total number of vehicles that would otherwise be in service would be greater. Service cuts on bus routes for fleet availability would not have been needed, and the TTC would be scrambling to find operating dollars to run a full complement of streetcar service plus all of the buses it has available.

Updated January 23, 2017 at 1:00 pm: The 95 York Mills restructuring was omitted in error by me in the original version of this post and the linked spreadsheet. This has been corrected.

The February 2017 schedules bring a relatively small set of changes to TTC operations.

Among them are a number of route restructurings where service is shuffled between the various local and express, where it exists, branches.

• 199 Finch Rocket will see:
  • more frequent AM peak service on the 199B STC to York University branch, offset by reductions on the other two branches;
  • more frequent midday service on the 199B offset by reduced service on the 199A to Finch Station;
  • slightly more frequent PM peak service on the 199A and 199B services offset by a reduction in service on the 199C to Morningside Heights.
• 191 Highway 27 Rocket will see improved peak service on the 191C branch to Humber College.
• 72 Pape’s peak period short turn service 72A to Eastern Avenue extended to become 72C to Commissioners with headway widenings to compensate on both the 72B Union Station and (now) 72C short turn.
• 24 Victoria Park will see peak period improvement in the 24B Consumers Road branch offset by a slight reduction in 24A to Steeles.
• 95 York Mills will see better express service to UTSC in the PM peak offset by a reduction on the local branches.

Other significant route changes include:

• 511 Bathurst bus operation will be changed so that on weekday peaks and midday half of the service will short turn at King Street. This will reduce the cost of operating the replacement bus service on the streetcar route, and will also reduce the peak bus requirement.
• 52 Lawrence West will be modified so that the 52F service to Royal York loops via Braecrest and returns east on The Westway rather than running north on Royal York and then east on Dixon Road. Schedules at various times will be modified to better reflect actual operating conditions. There will be a slight decrease in PM peak service levels on all branches, with a small increase in the early evenings weekdays.

Service improvements (beyond those involved in the items above):

• 108 Downsview (Midday weekdays)
• Various school trips have been rescheduled to better match dismissal times for students

Service trims on various routes continue to reduce vehicle requirements and costs for 2017:

• 511 Bathurst (as noted above)
• 504 King (PM peak)
• 505 Dundas (Midday weekdays)
• 26 Dupont (PM peak)
• 15 Evans (PM peak)
• 46 Martin Grove (PM peak)
• 37 Islington (Late Sunday evening 37C service on Steeles to Kipling cut back as 37B to Islington)
• Some operational changes have been made to hook up runs on various routes and address situations where a bus runs into a garage at roughly the same time as another one is scheduled to leave.

There is no announced date for the return of streetcars to 511 Bathurst. This depends on the timing of new streetcar deliveries from Bombardier and the condition of the remaining streetcar fleet. Although some old cars are in the shops for major rehabilitation, many more remain on the street with declining reliability.

Once 509 Harbourfront and 514 Cherry are fully equipped with new cars, the next route to be changed over will be 505 Dundas and then 511 Bathurst. If Bombardier deliveries continue at the late 2016 rate and do not ramp up until April 2017 (as planned), the conversion of 505 Dundas will not start until sometime in the second quarter.

Details are in the linked spreadsheet:

2017.02.12_Service Changes_v2

Regular readers will be familiar with many analyses published here that review the behaviour of transit service on streetcar and bus routes. One of the more galling parts of a transit rider’s life is the uneven headways (the times between vehicles) for almost all services that make the anticipated wait longer than the advertised average in the schedule.

This article is an attempt to devise a measure of this problem that can be used to show the degree to which actual service deviates from the scheduled value in a way directly related to rider experience. Some of the material is a tad technical, but not overly so. Also, readers should note that this is a first cut, and suggestions for improvement are welcome.

Wait Times Versus Headways

In theory, service is scheduled to arrive at a stop every “N” minutes, like clockwork. Therefore, the average rider will wait half of this time for a vehicle to show up. If buses are supposed to arrive every 5 minutes, then the average wait is 2.5 minutes. But the situation is more subtle than this.

If riders arrive at the stop at a uniform rate, say one each minute, they will not all wait the same length of time. Someone who arrives in the first minute will wait almost the entire headway, 4.5 minutes on average, while someone who arrives within a minute of the bus will wait on average only 0.5 minutes. The five riders who arrived in the five minute headway will have varying wait times, but on average the value will be 2.5 minutes per rider, and a total of 12.5 for the group overall.

    Rider  Wait Time
      1       4.5
      2       3.5
      3       2.5
      4       1.5
      5       0.5
    Total    12.5

The total above is the sum of a simple arithmetic series, and because of the values chosen here, the formula for calculating the total is quite simple. It is the square of the headway in minutes divided by two.

    Total = (N * N) / 2

If the service is “well behaved” with consistent headways, this is just a complicated way of getting to the same result with the average wait being half a headway (5 minutes divided by 2 giving 2.5 minutes). However, when headways are not consistent, the square factor in that formula shows its effect.

Suppose that vehicles are supposed to arrive every 5 minutes, but in fact two vehicles arrive sometime within a 10 minute window on varying headways. These might both be 5 minutes, but they could also be 6 and 4, 7 and 3, etc. In this case, the wait times behave differently. The table below shows how the average wait time for the ten riders accumulating at a stop during a 10 minute period vary depending on the regularity, or not, of the headways.

    First     Wait    Second    Wait    Total    Average
    Headway   Time    Headway   Time     Time     Wait
      5'      12.5'     5'      12.5'    25.0'     2.5'
      6'      18.0'     4'       8.0'    26.0'     2.6'
      7'      24.5'     3'       4.5'    29.0'     2.9' 
      8'      32.0'     2'       2.0'    34.0'     3.4'
      9'      40.5'     1'       0.5'    41.0'     4.1'
     10'      50.0'     0'       0.0'    50.0'     5.0'

It is self-evident that if two buses arrive every 10 minutes, the average wait time is 5 minutes, but this table shows how the values rise for different levels of headway inconsistency in between.

The TTC considers that vehicles that are only slightly off schedule (one minute early or up to five minutes late) are “on time” for purposes of reporting service reliability. The problem with this scheme is that the leeway it allows is very broad for routes that operate frequent service. Indeed, if a bus is planned to come every 5 minutes, two could come every 10 and the service would still be “on time”.

There is also the problem that riders on such routes don’t care about the schedule because it really is meaningless. They only care about reliable service, and a six-minute swing of “on time” values is often wider than the scheduled headway. The result is a meaningless metric of “on time performance”.

For wider scheduled headways, that six minute swing does not represent, proportionately, as much of a potential change in average wait times. Consider pairs of buses on a 20 minute headway:

    First     Wait    Second    Wait    Total    Average
    Headway   Time    Headway   Time     Time     Wait
      20'     200.0'    20'     200.0'   400.0    10.00'
      21'     220.5'    19'     180.5'   401.0'   10.03'
      22'     242.0'    18'     162.0'   404.0'   10.10' 
      23'     264.5'    17'     144.5'   409.0'   10.25'
      24'     288.0'    16'     128.0'   416.0'   10.40'
      25'     312.5'    15'     112.5'   425.0'   10.63'

Although the headways are wider, the percentage in change relative to the scheduled value is smaller, and so both the total rider-wait time and the average wait time don’t shift as much as they do for more frequent service for the same latitude in headway adherence. To get the same effect, proportionately, as in the first example, the two buses would have to arrive together in a 40 minute gap.

Note that these values behave the same way regardless of the assumed arrival rate of passengers provided that this rate does not change. The total rider-wait times would be higher for an arrival rate above one per minute, but the average values would remain the same.

There is also a presumption here that for infrequent services, riders will arrive uniformly over the headway. Where the route is the origin for a trip, riders might be expected to time their trips to minimize waits, although this behaviour can be thrown off if service is not reliable, especially if it is often early. For riders arriving at a bus-to-bus connection or at a subway-to-bus transfer, they are unlikely to be able to fine-tune their trips to just catch a bus. (The exception would be a network with protected, timed transfers, but the TTC does not have any of these.)

Crowding Effects

These values also affect on-board crowding because a vehicle carrying a wider than scheduled headway will accumulate more passengers and will have longer stop service times. It will become later and later, and the vehicle behind will eventually catch up. As with “average” wait times, the “average” crowding level will not represent the actual situation experienced by the “average” rider. Just as more riders wait longer than the planned average when buses do not arrive regularly, more riders are on the more crowded vehicles.

The problem is further compounded because most riders try to get on the first vehicle that arrives lest the second one be short-turned and they are left in a big gap waiting to continue their journeys.

If one were to poll 100 riders distributed between two streetcars, they would not complain about crowding if they were equally distributed and all had a seat. However, if 80 of them were on the first car and only 20 on the second, the average perception of crowding would be that the service was overloaded and that it did not show up reliably.

This is a fundamental split between service as the TTC sees it (on average) and as riders see it (as an individual experience).

Measuring the Difference Between Actual and Scheduled Wait Times

In preparing this article, I have been wrestling with various ways to calculate and display these values in a useful way. There are pitfalls both in the methodology and in the charting of information. The examples below are only one way to present this information, and should not be read as definitive.

Here is the basic premise:

• Within each hour of the day, vehicles pass by a location on a route, each of them on its own headway.
• From the headway values, one can calculate the wait time for riders assuming an arrival rate of one/minute. The total of these times divided by 60 gives the average wait per rider over the hour period.
• The count of vehicles within each hour is easily converted to an average scheduled wait time of one half the average headway.

These values are graphed for the 504 King route for Tuesday, November 1, 2016. A sample page is shown below and the full sets are linked as PDFs.

504_20161101_avgwaittime_wb

504_20161101_avgwaittime_eb

In these charts the solid lines represent the rider wait time in minutes/passenger while the dotted lines are the average that would have applied if all of the service had been evenly spaced over an hour. Each colour shows data for a different hour through the day, and the horizontal positions show various locations along the route. Note that the rider wait times are generally higher than the average times implied by the vehicles/hour.

There are several caveats about this presentation:

• For the 6:00 am line (red on page 1 of each set), the first headway is assumed to occur from 6:00 am until the arrival of the first vehicle. This will generally understate the headway on which that vehicle is operating.
• Depending on how badly disrupted the service is, the wait time contributed by the first vehicle passing in any hour may include time accumulated during the previous hour. This causes wait time to be mis-attributed, and it is possible for the rider wait time in the previous hour to be understated. For example, if one car passes a stop at 7:56 and the next one at 8:04, all of the rider wait time associated with the second vehicle will be included in the 8:00 data as will the vehicle in the total count. This effect is generally small enough that it balances out, but it can be more pronounced on wider headways where there are fewer vehicles to smooth out the data.
• In some cases, a gradual increase in rider wait times is visible along the length of the route showing how service becomes progressively more bunched into pairs.
• Major delays can result in a large increase in rider wait time because the long gap dominates the calculation, and following cars contribute almost nothing to this value. However, the number of cars/hour could still be at the scheduled level and the “average” wait times are not affected.

For reference, I have also included here the chart of overall service for the day.

504_20161101_chart

There is a disruption of service westbound at Dufferin between 10:00 and 10:30 am. This causes headways west of this point to be bunched, and that bunching is reflected back on the eastward trip from Dundas West as a parade of cars makes its way across the city. Note that nothing was short-turned into the gap eastbound, and it was not until the next westbound trips after 11:00 that service began to be sorted out.

A delay eastbound at Queen & Roncesvalles just after 15:30 shows up as a spike in “3 pm” rider wait times at that location, but most of the gap passes points further east during the “4 pm” interval. This explains the behaviour of the plots for these two time periods.

As I wrote above, this is a first cut at attempting to measure what riders experience versus what the TTC will typically state as its quality of service.

A useful corollary to this would be to examine data from automatic passenger counters on the vehicles, but these have not yet been rolled out fleet wide. This would allow us to tie the service reliability values to crowding conditions on vehicles.

What will not show up, of course, is the number of passengers who simply never get on because they tire of waiting for a vehicle with room on it. At least with detailed counts, we would know how often “full vehicles” actually occur, and especially cases of lightly-loaded ones that are not pulling their weight because they are part of a parade. That is a task for a future round.

If I get any other ideas about how to present this information, I will update this article with additional examples.