GPS Enabled Identification of Cruising for Parking

Rachel Weinberger, PhD

Regional Plan Association

One Whitehall Street

New York, NY

Adam Millard-Ball, PhD

UCLA

Robert Hampshire, PhD

USDOT

Ellis Calvin

Regional Plan Association

Carlos Mandeville

Regional Plan Association

Tayo Fabusuyi, PhD

Transportation Research Institute

University of Michigan

Ann Arbor, MI 48109

Fabusuyi@umich.edu

Michelle Neuner, PE

Leidos

1750 Presidents Street
Reston, VA 20190

Michelle.L.Neuner@leidos.com

Allen Greenberg*

Federal Highway Administration

1200 New Jersey Ave., SE
HOTM-1

Washington, DC  20590

Allen.Greenberg@dot.gov

Word Count: 3,485 + 11 tables/figures x 250 words (each) = 6,235 words

^*The views expressed in this research are not necessarily those of the Federal Highway Administration.

Introduction

Cruising for parking is a serious and vexing problem for cities across the world.  Under a U.S. Department of Transportation (USDOT) Small Business Innovative Research grant, Weinberger, Millard-Ball and Hampshire developed a strategy to examine cruising for parking using Global Positioning Systems (GPS) traces to see what strategies drivers employ in their parking search (14).  The original work utilized some very controlled datasets –a set of GPS traces accompanied by video that had been collected for unrelated research in Ann Arbor, the California Household Travel Survey, and an “uncontrolled” set of vehicle GPS traces from a data consolidator.  

The original work was to develop a tool by which policy makers could identify cruising hotspots and to better understand the geographic and temporal extent of parking search.  The findings were reported as provisional as the intent of the work had been proof of concept rather than in pursuit of particular behavioral understanding or policy outcomes.  Nevertheless, top line findings indicate that cruising may be a smaller problem than often thought. In San Francisco and Ann Arbor, cruising was a component of approximately 6%-7% of automobile trips as opposed to the often cited number of 30% (reference to Shoup), even though the 30% figure, derived by averaging a variety of small parking cruising studies over a period of many years in specific areas where cruising was known to be problematic, is often mispresented. Despite its generally low prevalence, high levels of cruising were found in some instances.

The research presented here is an application and generalization of that work to two cities; it is part of a larger effort to apply the tool to disparate urban contexts.  For the analysis we selected two cities that were known as early adopters of performance parking –the strategy by which cities expect to reduce cruising by setting occupancy targets for their [metered] streets and then set parking meter rates to meet those targets—or other innovative parking and curb management strategies.  

Additional features of the work include an assessment of before and after policy interventions and different cruising dynamics across areas distinguished by different features.  Specifically, we look at:

• Seattle, Washington, and examine travel behavior in a pre-COVID rate adjustment; we also look at what happens with cruising during the COVID pandemic both when the meters were turned off entirely (April 4, 2020) and then when they were turned on again (July 13, 2020).

• Washington, DC, during a time period that is strictly prior to the COVID pandemic with a particular eye to understanding cruising in areas around Metrorail stations compared with other parts of the city and areas around the city’s three stadia on event and non-event days.

It is our intention with this work to test different available datasets and to develop increasingly accessible ways for cities to perform their own analyses to understand cruising within their city limits.  Depending on data availability, cities can understand cruising at any geography and time period allowing them to develop strategic, targeted interventions.

Literature Review

Popular accounts and news reports often paint a dismal picture of parking availability in urban centers. The same is true of surveys that ask drivers to self-report parking search times. In one district of Brisbane, Australia, for example, respondents reported an average search time of 13 to 16 minutes (1), while in San Francisco a study found averages of 6-7 minutes (2). 

Studies that rely on driver perceptions and self-reported times, however, may be problematic if drivers find it easier to recall instances when parking was difficult to find. Other methods have been developed to measure cruising more objectively, such as following cars with bicycles (3), undertaking test runs to search for parking along pre-defined routes (4), or recording how many cars pass a vacant space (5). Simulations have also been used (6, 7). Across research approaches, studies commonly risk overestimating the prevalence of cruising. First, there may be a selection bias in that studies tend to focus on places and times of day where parking availability is known to be low. Second, most studies do not distinguish between parking search and excess travel (which we call cruising). As we discuss elsewhere (8), drivers might start their parking search prior to their destination, and even stop short of their destination if they find a space—in which case excess travel and cruising would be negative. In a place with no through traffic, half of the traffic will be searching for parking (those arriving rather than leaving), but there might be little or no cruising.

Indeed, some studies suggest that the overall prevalence of cruising is fairly low. Household travel surveys can overcome the selection bias, and analysis of the Dutch National Travel Survey yields an average cruising time of just 36 seconds per trip (9), even though these data are self-reported and thus perceptions might still be an issue. More recent studies use GPS trace data to overcome all three challenges—perceptions, selection bias, and excess travel. In Greece, the mean search time was 131 seconds with 75% of drivers searching for less than 2.5 minutes, but these figures only include drivers who searched for on-street parking (10). In San Francisco and Ann Arbor, our own work using GPS traces finds a similar mean search time of 117 – 168 seconds. However, because only 5% to 9% of trips involve any cruising, parking search accounts for less than 0.5% of private car traffic (11). We suggest that cruising is partly constrained by drivers willingness to accept a longer walk where they know available parking to be scarce, and as a result parking short of their destination if they pass a vacant space (8).

Most empirical studies to date have focused on assessing the overall extent of cruising. Only a few have examined the effectiveness of policy interventions to reduce cruising, such as Alemi et al. (4) and Millard-Ball et al. (12) who document the success of parking pricing and demand-based meter price adjustments in reducing cruising in San Francisco. There is also little research that analyzes how the built environment and other factors affect the level of cruising. One exception finds that non-residential land uses and higher density neighborhoods are associated with more cruising, as are trips at certain times of day and of certain lengths(10). 

Data and Methodology

The system we developed matches GPS traces to a street network and then compares the path a vehicle has traveled with the shortest path available to the driver’s final geographic reference point.  We state the end of their trip in these vague terms to emphasize that we know only the vehicle’s last geographic reference point, but we do not know the driver’s actual destination.  In a GPS augmented travel survey that distinction could be clarified but our method relies on a much greater volume and density of information than could be assembled in a travel survey. A detailed description of the map-matching algorithm is available in prior work (13).

Once the traces are matched to the network and a comparison is made, we are able to determine whether the trip includes a “cruising” component.  This is defined as a path that is at least 200 meters greater than the shortest path. Finally, we match the cruising and non-cruising trips to their geographies and we are able to map the areas where cruising occurs.  A more detailed discussion of the process is documented in the report Parking-Cruising Caused Congestion (14).

We reviewed a variety of possible data sources including consolidated data from navigation devices, app-based location data that is not specifically from navigation systems, GPS augmented household travel surveys, and cellphone-based location data.  For this project, we selected app-based location data and navigation-based location data.  In this paper we report only on the analysis from navigation data.  Ideally for this research individual vehicle traces would be directly available to the analyst; individual traces were used in developing the system.  The reality of privacy concerns and data availability suggested we pursue a different strategy in this part of the project.  A private vendor, StreetLight, aggregates data from GPS navigation devices, but their data protection protocols preclude them from releasing the data. In a partnership agreement, StreetLight analyzed the data for both Seattle and Washington, DC using our cruising identification algorithms. 

For the current analysis we defined 15,000 unique geographies in each city, and StreetLight reported the results of the analysis aggregated to those geographies.  The geographies are a combination of individual streets and census block groups.

In Seattle, we were able to study geographic, temporal, and policy impacts.   The Seattle case, illustrated in Figure 1, allows us to look at 280,000 trips on 2,100  individual streets that were characterized as having parking meters, 11,400  streets surrounding metered areas (we assume a policy change on a metered block will have knock-on effects in the surrounding area), and finally the remaining parts of the city were studied in block groups (for cost efficiency).  We selected three policy events and six corresponding time frames as shown in Table 1. We analyzed cruising before and after each policy event.  

Policy event/date	Before dates	After dates
Business as usual rate adjustment – January 28, 2020	1/6/20-1/27/20	2/5/20-2/28/20
COVID Meter Shut off – April 4, 2020	3/21/20-4/3/20	4/05/20-4/18/20
COVID meter reinstated at $0.50/hour – July 13, 2020	6/29/20-7/12/20	7/14/20-7/27/20

Table 1 Seattle Policy events

Figure 1 Seattle, Washington selected geographies Meter, Near Meter, and Un-metered locations

Washington, DC data were analyzed for the calendar year 2018.  Except in the Penn Quarter/Chinatown area where meter rates were regularly adjusted in accordance with predefined rules for a performance parking pilot project, and for which results have been documented elsewhere (get Soumya paper), there were no significant policy events applying to parking meters.  Instead, we chose to look at implicit policies cross-sectionally by contrasting streets around Metrorail stations and around the three event venue/sports stadia on event and non-event days and “all other areas.”  We looked at xxx individual streets in station areas, zzz streets around Nationals Park (baseball stadium), Audi Field (soccer stadium), and Capital One Arena.  These geographies are illustrated in Figure 2.

Figure 2 Washington, DC selected geographies, Metro Stations, event venues.                        

A general caveat with respect to the outcomes, given the aggregated nature of the data, is that we know only the geography where the trip ended and not the streets on which cruising may have occurred.  The distinction matters when thinking about the number of cruising trips that end on metered blocks for example.  It would be easy to mis-construe that finding to think that it means there is more cruising on metered streets when, in fact, a driver, after failing to find free parking, will take a space on a metered block because it is on the metered blocks that parking is available. 

Results

Our analysis is first done at the block group level (we aggregated the street level data to block groups), and then we are able to drill down to the block level for our geographies of particular interest.  The variables of primary interest are where cruising trips occur and the average time spent cruising.

Seattle

In Seattle, approximately 7% of trips include a cruising component, and the average time spent cruising was just over one minute.  Trips that end on metered blocks have a cruise rate of 9.6% and average time cruising is 122 seconds.

Interestingly, while cruising trips are relatively concentrated in the downtown, the amount of time people spend looking for parking is fairly uniform across the city (see Figure 3).

Figure 3

Seattle Policy Case Studies

The first case is a business-as-usual performance-based meter price change. On January 28, 2020, Seattle Department of Transportation (SDOT) implemented a routine meter rate adjustment consistent with their meter performance goals.  Depending on average occupancy, in some areas the hourly meter price increased by $0.50, in some it decreased by that amount, and in the areas that were meeting performance goals the meter rates remained unchanged. 

Our analysis shows that in the areas where meter prices were lowered, there was a slight decrease in tripmakers finding a parking spot on metered streets.  This could reflect lower parking availability due to the rate adjustment.  We see an increase in trips that involve cruising ending on metered blocks in areas where the meter price was increased.  This could reflect a greater pressure on the surrounding area and more availability on the metered blocks.  As expected, cruising in non-metered areas does not change. Cruising ending on blocks where the price did not change is also relatively stable.  These findings are illustrated in Figure 4, Average cruising frequency before and after price changes.

The amount of time drivers spent looking for parking decreased in all cases, suggesting a successful policy implementation (Figure 5, Average Cruising Time).  The most notable observation after the meter price change is that cruising times decreased in areas and times of day where meter prices decreased. This could suggest that, given availability, drivers were willing to more quickly take a metered parking spot at a lower price.  In places and times where meter prices increased, cruising frequency increased only slightly, while cruising times decreased slightly.  We do not have a theory to explain this finding.

COVID-19 pandemic meter pricing changes

We turn our attention to two COVID-19 policy responses.  In response to the Covid-19 pandemic and lockdown measures implemented by the city and state, trips were largely limited to only the most essential. To ease the burden on essential travel, SDOT suspended all paid street parking on April 4, 2020. As businesses reopened, SDOT reinstated paid street parking at a citywide rate of $0.50 to allow for more reliable access to the curb. 

The pandemic dramatically changed travel patterns, while parking policy changes appear to have had little impact on overall trips. In Period A, which takes place entirely before the lockdown measures in response to the COVID-19 outbreak, trips in Seattle follow the expected diurnal pattern of morning and evening rush hour peaks, with slightly more trips in the After period. In Periods B and C, during the first few months of the pandemic, the total number of trips was much fewer, and tended to peak in the afternoon.

Though trip-making in April dropped to a third of what it had been prior to the COVID-19 pandemic and was still quite low in July when meters were made operational again, and while the diurnal pattern of trips was very different, the pattern of cruising remained curiously consistent.  The reasons for this may include less turnover in parking spaces, or a preference for street parking over more expensive garage parking.

Figure 6  *Period A= Business-as-usual  Period B=meters are turned off  Period C= meters are reinstated at $0.50 across all areas

Washington, DC

In Washington, DC, we studied the difference in cruising patterns around Metrorail stations versus the rest of the city, and we looked at cruising for parking on stadium event versus non-event days.  Figure 2 shows the areas around Metrorail stations and the three major stadia for which we have block level data and those where we have data aggregated to census block groups.

In the baseline, approximately 6% of trips were classified as cruising.  The average amount of time spent cruising is just over 2 minutes.

Differences from the baseline that are most pronounced include altogether higher rates of cruising around Metrorail stations –this is likely a function of density; a higher rate of cruising for parking on weekends accompanied by longer search times –likely due to the fact that meters are not in effect on Sundays and could be exacerbated by reduced Metrorail service, and, in turn, ridership, on the weekends.

Stadium Events

As expected, cruising around stadiums on non-event days does not differ from cruising in other parts of the city but cruising around stadiums on event days is higher, effecting 6% of trips.

In particular, Audi Field, which is the least accessible of the stadia by Metrorail, shows the most pronounced differences. As illustrated in the first panel of Figure 8, the diurnal distribution of cruising trips outside the stadium catchment area on event and non-event days tracks very closely with the distribution of cruising in the stadium catchment area on non-event days.  The greater fluctuation in the line representing non-event days in the stadium catchment area is due to some extent to low data fidelity.  Nevertheless, cruising for parking, both early in the day and toward the afternoon on event days, is markedly higher than was found in the other times and places. 

Figure 8

Metrorail Station Areas

Conclusions and further research

The research presented here shows the range of possibilities for studying cruising dynamics under a variety of implicit and explicit policy scenarios.  The Seattle price changes are explicit policy changes, and the study of cruising around different land uses and levels of transit accessibility form implicit policy distinctions.  In addition, the massive reduction of travel due to the Covid-19 pandemic allowed us to see what happens with cruising in a reduced tripmaking context.  The use of GPS obviates the need forwide-ranging, expensive field data collection.  The broad contextual view, due to the comprehensive coverage, allows us to identify otherwise unexpected impacts of cruising and of policies to combat it.

Some of the findings seem counter-intuitive or even paradoxical: a meter price increase in Seattle led to an increase in cruising but for shorter search times; cruising around transit stops is greater than in other parts of the city –suggesting there is a tolerance or acceptance of certain levels of cruising.  Under the scenario where trips were reduced to a third of normal, cruising patterns remained the same.  The latter finding suggests that cruising may be self-limiting and that there is some degree of tolerance for cruising regardless of the circumstances.

Other findings are clearly as most would predict –spikes in cruising around Audi Field in Washington, DC, on event days and more cruising on weekends –when meters are not operational—compared with weekdays when they are, being two examples.  

The particular findings in Seattle and Washington, DC, may not apply to other cities as each city is unique with respect to density, destinations, transit infrastructure, and land use patterns, not to mention parking and meter policies, but the applicability of the system to look closely across a region should hold some appeal.

Next steps

In some respects, the findings raise as many questions as they answer.  The next phase of this research will test other possible data sources.  In particular, we will compare the navigation based aggregated results from Seattle with GPS points that are derived from non-navigation apps.  The disadvantage will be in data fidelity, but the potential advantage of the individual traces will be in better understanding the micro-level cruising dynamics.  

We will continue to analyze cities for the purpose of gaining increasing insight into questions that inform curb management and increase our understanding of how cruising can be solved –if at all.

Under this Federal Highway Administration (FHWA) research project, a final report and presentation materials will be developed. The parking cruising assessment methodology will be made available as open source code. A users’ manual/explanation of the methodology will also be developed.

References

1.         Lee, J. (Brian), D. Agdas, and D. Baker. Cruising for Parking. New Empirical Evidence and Influential Factors on Cruising Time. Journal of Transport and Land Use, Vol. 10, No. 1, 2017, pp. 931–943. https://doi.org/10.5198/jtlu.2017.1142.

2.         SFMTA. SFpark Pilot Project Evaluation. San Francisco Municipal Transportation Agency, San Francisco, 2014.

3.         Shoup, D. C. The High Cost of Free Parking. Planners Press, Washington, DC, 2005.

4.         Alemi, F., C. Rodier, and C. Drake. Cruising and On-Street Parking Pricing: A Difference-in-Difference Analysis of Measured Parking Search Time and Distance in San Francisco. Transportation Research Part A: Policy and Practice, Vol. 111, 2018, pp. 187–198. https://doi.org/10.1016/j.tra.2018.03.007.

5.         Hampshire, R. C., and D. Shoup. What Share of Traffic Is Cruising for Parking? Journal of Transport Economics and Policy, Vol. 52, No. 3, 2018, pp. 184–201.

6.         Arnott, R., and P. Williams. Cruising for Parking around a Circle. Transportation Research Part B: Methodological, Vol. 104, 2017, pp. 357–375. https://doi.org/10.1016/j.trb.2017.07.009.

7.         Levy, N., K. Martens, and I. Benenson. Exploring Cruising Using Agent-Based and Analytical Models of Parking. Transportmetrica A: Transport Science, Vol. 9, No. 9, 2013, pp. 773–797. https://doi.org/10.1080/18128602.2012.664575.

8.         Millard-Ball, A., R. C. Hampshire, and R. R. Weinberger. Parking Behaviour: The Curious Lack of Cruising for Parking in San Francisco. Land Use Policy, Vol. 91, 2020, p. 103918. https://doi.org/10.1016/j.landusepol.2019.03.031.

9.         van Ommeren, J. N., D. Wentink, and P. Rietveld. Empirical Evidence on Cruising for Parking. Transportation Research Part A: Policy and Practice, Vol. 46, 2012, pp. 123–130. https://doi.org/10.1016/j.tra.2011.09.011.

10.       Mantouka, E. G., P. Fafoutellis, and E. I. Vlahogianni. Deep Survival Analysis of Searching for On-Street Parking in Urban Areas. Transportation Research Part C: Emerging Technologies, Vol. 128, 2021, p. 103173. https://doi.org/10.1016/j.trc.2021.103173.

11.       Weinberger, R. R., A. Millard-Ball, and R. C. Hampshire. Parking Search Caused Congestion: Where’s All the Fuss? Transportation Research Part C: Emerging Technologies, Vol. 120, 2020, p. 102781. https://doi.org/10.1016/j.trc.2020.102781.

12.       Millard-Ball, A., R. Weinberger, and R. Hampshire. Is the Curb 80% Full or 20% Empty? Assessing the Impacts of San Francisco’s Parking Pricing Experiment. Transportation Research Part A: Policy and Practice, Vol. 63, 2014, pp. 76–92. https://doi.org/10.1016/j.tra.2014.02.016.

13. Millard-Ball, A., Hampshire, R.C., and Weinberger, R.R. (2019). “Map-matching poor-quality GPS data in urban environments: the pgMapMatch package.” Transportation Planning and Technology, 42(6), 539-553.

14.  Weinberger, R.R., Millard-Ball, A., and Hampshire, R.C. (2016). Parking-Cruising Caused Congestion. March 1. Available online at:  <https://ssrn.com/abstract=2906528> or <http://dx.doi.org/10.2139/ssrn.2906528>.