Final Fall Quarter Report The Traffic Annoyance Effect and Other Predictors for Shuttle Ridership S Leea☨, K Phanb☨, J Zhaoc☨, D Ouyangc, and G Katzc a
Department of Materials Science and Engineering, Stanford University b Department of Computer Science, Stanford University c Department of Civil and Environmental Engineering, Stanford University d Department of Management Science and Engineering, Stanford University ☨ These authors contributed equally to this report Abstract This report to the Managers’ Mobility Partnership explores ways to reduce single occupancy vehicle rates for commuters living three to ten miles from major employers. In interviews, partners were concerned about a “traffic annoyance effect”, where commuters may be discouraged from taking shuttles if those shuttles would be stuck in the same traffic as single occupancy vehicles. We found no significant statistical correlation between traffic delay, either absolute or as a percentage of journey time, with ridership in the Palo Alto Crosstown, Marguerite AE-F, or Marguerite U lines for all departure times and just peak hour departure times. This suggests that for intra regional shuttles, the traffic annoyance effect is not a significant determinant of ridership, so right of way lanes for local shuttles may not substantially increase ridership by decreasing the traffic annoyance deterrent. Keywords local congestion, municipal shuttles, ridership, traffic delay, mid distance commuting, commuter behavior
1. Motivation and Scope In our midterm report, we presented to three options domain areas to the partners to consider for our project: linking intercity traffic data, analyzing best practices for partner run shuttle services, and exploring mobility access for communities at risk. Multiple partners expressed an interest in the intersection of the traffic and shuttle domain areas, and also expressed an interest in intra-regional commuting. In that vein, our research questions is to investigate ways to reduce single occupancy vehicle for commutes of less than 10 miles from a large employer. For longer distances, employer operated shuttles, CalTrains, and the state highway system are all outside the jurisdiction of the
municipalities in the Managers’ Mobility Partnership, and for shorter distances, the Stanford Public Policy Practicum is focusing more on bicycle routes and policy. So we are focusing on shuttle ridership vs. single occupancy vehicles because the decision between these two options is most relevant for commuters who live between 3-10 miles of their employer, and contribute most significantly to local municipal road congestion. Commute Distance
Relevant Modes
Relevant Partners
Study Groups
< 3 miles
Walking, Biking
Major Employers , Municipal ities
Public Policy Practicum
Lee, Phan, Zhao et al, Sustainable Urban Systems Projects (2016) 2 3-10 miles
<10 mi
Shuttle, SOV
Shuttle, SOV, CalTrains
Major Employers , Municipal ities
Sustainabl e Urban Systems Team
Major Employers , Regional Transit Agencies, CalTrans
Solutions already implement ed
Table 1 Selection of commuting distance scope, selected scope for this project (3-10 miles) in italics
In this presentation and future quarters, we hope to assess potential predictors for shuttle ridership. As of now, we have received fully disaggregated ridership boarding data by stop, date, and time for the Marguerite AE-F, U, and Palo Alto Crosstown shuttles, so our analysis begins with these two lines. Multiple partners expressed concern about a traffic annoyance effect: that commuters were much less likely to ride a shuttle if that shuttle was also stuck in the same traffic that a private car would experience. An assessment of the extent of this traffic annoyance effect could provide substantial motivation for a fully dedicated or part time Bus Right of Way Lane on municipal streets.1 However, the exact extent of this effect for the shuttle services operated by the partners remains unquantified. 2. Methodology and Analysis For the three shuttle lines for which we had sufficiently disaggregated ridership data, we used google maps to find the precise 1
ichler, M., & Daganzo, C. F. (2006). Bus lanes E with intermittent priority: Strategy formulae and an evaluation. Transportation Research Part B: Methodological, 40(9), 731-744.
coordinates for each stop, and mapped them using Python. We then used the google maps API to figure out the time it took for a vehicle to take the route specified at the time that the shuttle was actually scheduled, as well as an ideal no traffic travel time. In this way, we were able to have a quantitative approximation of traffic annoyance. In order to ensure that different sized routes do not have disproportionate impact, we created two measures of traffic annoyance: one was the absolute time difference between the actual route and the ideal, traffic free route and the second was the time delay as a fraction of the traffic free route time: Annoyanceabsolute ≡ T traffic − T no traffic Annoyancepercent ≡
Stop
T traffic−T no traffic T no traffic
× 100 %
Address
Coordinates
Station
95 University Ave, Palo Alto, CA 94301
37.4434248,-12 2.1673629
Avenidas
450 Bryant St, Palo Alto, CA 94301
37.4461182,-12 2.1646053
Lytton Gardens 649 University Ave, Palo Alto, CA 94301
37.4503357,-12 2.1600214
Channing House 850 Webster St, 37.4464337,-12 Palo Alto, CA 2.1561567 94301 Rinconada Library
1213 Newell Rd, Palo Alto, CA 94303
37.4449696,-12 2.1413488
Palo Alto Art Center
1313 Newell Rd, Palo Alto, CA 94303
37.4439836,-12 2.1410249
Jordan Middle
750 N.
37.438499,-122.
Lee, Phan, Zhao et al, Sustainable Urban Systems Projects (2016) 3 School
California Ave Palo Alto, CA 94303
1368917
Midfield 2775 37.4332772,-12 Shopping Center Middlefield Rd, 2.1308812 Palo Alto, CA 94306 JLS Middle School
480 E Meadow Dr, Palo Alto, CA 94306
37.4225669,-12 2.1193003
Mitchell Park Community Center
3700 37.4223141,-12 Middlefield Rd, 2.1148408 Palo Alto, CA 94303
Stevenson House
455 E Charleston Rd, Palo Alto, CA 94306
37.4190945,-12 2.1155561
Table 2 Coordinates for Palo Alto Crosstown Shuttle
In order to quantify the effect of annoyance on ridership while controlling for different route sizes, we took the total boardings for each route’s departure time and divided it by the average total boardings for all departure times on that route: ∑
∑ boardings(Departure Timei)
⋁ stops Ridership F actor1 ≡ ∑ boardings(Departure T ime1) ÷ ⋁ Departure times number of departure times
⋁stops
In order to see if this effect was only prevalent during commute hours, we performed the correlation between annoyance and ridership factors with all the points, and just the points from 7:00-9:00 am and 3:00-6:00 pm. To make this easier to visualize in future tests where we may want to asses the effect on passengers at boarding or disembarkation, we combined the roads where traffic was predicted, boarding numbers, and route data in a single visualization. With more layers, this visualization has the potential to help us spot patterns in the movement of people, vehicles, or the intersection between the two from which we may formulate alternative ridership predictors.
Figure 1 The Palo Alto Crosstown Map (blue) with boarding data (radius of the red circles), stops (in green), and traffic rates (road colors).
Finally, in order to run a preliminary analysis of other location based predictors, like the distance from home to shuttle stop, we used the US Census’s OnTheMap tool to map the percentage of incoming commuters in Palo Alto, Stanford, Menlo Park, Redwood City, and Mountain View that come from those same cities. This also gives us a sense of the overall demand for local transportation options. 3. Results and Discussion No significant correlation was found between the ridership factor and the traffic delay, either in absolute time or in
Lee, Phan, Zhao et al, Sustainable Urban Systems Projects (2016) 4 percentage of the no traffic route, as can be seen in Table 3. Furthermore, the best fit lines had slopes of almost zero as can be seen in Figure 2, and this result was robust regardless of whether we considered all route times or just those during peak traffic, when the annoyance factor might logically be greater.
Dynamics survey has estimates of commuters to and from jurisdictions down to the census tract level. Partner
Intra Partnership Commuting
Redwood City
10.6 %
Menlo Park
17.3 %
Palo Alto
19.3 %
Stanford
16.0 %
Mountain View
13.2 %
Table 4 Percentage of workers in Partnership jurisdictions who live in Partnership jurisdictions
Figure 2 No effect pattern between ridership and traffic delay including and excluding off peak hour traffic and by absolute and percentage traffic delay measures.
All points
Rush hour only
Slope
P Value
Delay Time
-0.0068
0.87
Percent Delay
-0.0040
0.80
Delay Time
0.0020
0.97
Percent Delay
0.0079
0.73
Table 3 Lack of significant correlation between
ridership and traffic delay including and excluding off peak hour traffic and by absolute and percentage traffic delay measures.
In light of these results, we decided to explore alternative potential determinants for shuttle ridership. The US Census’s Longitudinal Employer-Household
If broken down into a more granular level, these origin data can be used to asses the effect of home to stop distance on shuttle ridership, and help in a connectivity analysis. However, this will only be meaningful if we have shuttle ridership data disaggregated by stop and time of day. 4. Conclusions and Future Work We found no evidence that on the scale of local intra city shuttles, time spent in traffic is a significant determinant of shuttle ridership. This finding was robust regardless of whether we used all the departure times, or just those during rush hour and an absolute measure of delay time or a percentage measure of delay time. However, we cannot conclude that this pattern holds true for all shuttle programs operated by the partners, because we lack sufficiently granular data. This shortcoming must be addressed if we are to test more potential determinants of shuttle ridership with the hope of being able to define precisely their generality and applicability to all members
Lee, Phan, Zhao et al, â&#x20AC;&#x2039;Sustainable Urban Systems Projects (2016) 5 of the Partnership. In the future, we hope to provide a more comprehensive list of potential correlates as well as expand the ridership dataset. This may include: 1. Distance from stop to work 2. Distance from home to stop 3. Connections (also connectivity analysis) 4. Inconsistency of travel time 5. Frequency of shuttles Primary surveys on declared preferences may be the most efficient way to assess some of these hypotheses. With more significant findings, we will be able to inform policy choices like the potential of right of way lanes to increase shuttle ridership, and possibly conduct an economic analyses on the relative value of time saved for commuters like that implemented by Eichler and Daganzo.