Treatments fall into those applied along the link and those applied at the junction. Link treatments are bus lanes in their several forms, kerbside, offset from the kerb outside a parking lane, or in the median, operating full-time or in peak periods only (NACTO, 2016). Where bus flows are too low to justify a permanent lane, intermittent bus lanes that are cleared of other traffic only as a bus approaches have been proposed and analysed (Viegas and Lu, 2004; Eichler and Daganzo, 2006). Junction treatments include queue jump lanes, which let a bus bypass the queue on a short approach lane and leave ahead of it on a special signal phase, and signal priority itself. Stop-related measures, such as wider stop spacing, kerb extensions and boarding through all doors, attack dwell time and the delay in re-entering traffic (Kittelson & Associates et al., 2007).

Transit signal priority alters signal timing in favour of an approaching bus, most often by extending a green or truncating a red. It differs from pre-emption in that the controller stays in coordination (Smith et al., 2005). Its benefit depends on detail. Priority is most effective where the stop is on the far side of the junction, because the arrival time of the bus at the stop line is then predictable. It imposes least cost on cross traffic when it is conditional, that is, granted only to buses running behind schedule, a strategy shown by Furth and Muller (2000) to improve punctuality with little disturbance. Skabardonis (2000) sets out the range of passive and active strategies and the circumstances that favour each.

Traffic on a corridor changes for many reasons unrelated to a scheme, so a simple comparison of bus speeds before and after is weak evidence. The stronger quasi-experimental designs add a comparison group and a series of observations on either side of the change (Shadish et al., 2002). Transit agencies are well placed to use them, since automatic vehicle location and passenger counting systems archive the running time of every trip and the load on every bus, and methods for turning those archives into measures of speed and reliability are established (Furth et al., 2006).

The first task is to divide observed bus travel time into its components by location. Archived location data, at stop level or finer, allow running time between stops, dwell time and, with care, signal delay to be separated for each segment and time of day (Furth et al., 2006). The variability of each component should be extracted along with its average, because passengers and schedulers are affected by unreliable trips at least as much as by slow ones. The output is a delay profile of the corridor: a table or strip map showing, for each junction and stop, how much time is lost and how consistently.

Objectives should be agreed before treatments are discussed. The procedure recommends stating them in terms of people and not vehicles: person-delay across all modes at each junction, bus running time and its spread, and schedule adherence. Vehicle-based level of service, as computed by conventional capacity analysis (Transportation Research Board, 2016), remains a necessary check on general traffic but gives a bus with forty passengers the same weight as a car with one. Fixing the measures at this stage also fixes what must be collected for the evaluation in Step 7.

With the delay profile and objectives in hand, each segment is assigned the lightest treatment that addresses its dominant source of delay. Table 1 sets out the pairing proposed here. The underlying principle, consistent with the published guidance (Danaher, 2010), is that a mid-frequency corridor seldom needs a continuous lane. Delay is usually concentrated at a few junctions, and a queue jump or a short length of lane on the approach captures most of the benefit at a fraction of the kerbside cost. A reserved lane over a longer length is kept for segments where queues regularly extend back through upstream junctions.

Signal priority and stop position should be settled in one exercise. Moving a stop to the far side, consolidating two closely spaced stops, and choosing between green extension and early green all interact, and treating them separately tends to produce priority requests that arrive at the wrong time. The mid-sized city has one advantage here. Because buses are not frequent, successive priority calls rarely conflict, and cross-street delay from an occasional extension is small. The design should set a limit on the extension, define the lateness threshold for conditional priority, and record the assumptions so that they can be revised in service.

Most objections to bus priority on a commercial street concern the kerb: parking, loading and access to premises. The procedure treats these as design inputs. Parking and loading activity should be surveyed by block and time of day, alternatives identified on side streets or in off-peak windows, and the rules for the bus lane (hours, permitted users, turning vehicles) written with enforcement in mind. A lane that is routinely blocked delivers little, and the means of enforcement, whether patrols or cameras where the law allows them, belong in the scheme from the start (NACTO, 2016).

Opening the corridor in stages serves both engineering and politics. Signal priority and stop changes can precede lane reallocation, since they require no road space, and a lane can be introduced with paint and signs before permanent works are committed. Each stage gives operating data against which the next can be adjusted. Staging does carry a risk, which is that an early, partial stage yields a modest gain and weakens support for the rest; the sequence should therefore begin with the junctions where Step 1 found the largest delays.

The evaluation uses the measures fixed in Step 2 and the same archived data sources as Step 1. The design proposed is an interrupted series with a comparison corridor: bus running times, their variability and passenger loads are extracted for a substantial period on either side of each stage, for the treated corridor and for a similar corridor left unchanged. Differences in the change between the two give a more credible estimate of effect than the raw change alone (Shadish et al., 2002). General traffic should be measured as well, and ridership interpreted cautiously, since patronage responds slowly and to many influences beyond travel time (Currie and Delbosc, 2011).