SMASH

In this post I (Zheng Liu) want to share some of my ongoing research that I recently presented at the AAG 2022 Annual Meeting.

Introduction

Recent studies on the changes in BSS mentioned there is an increase in the BSS trip duration during the COVID-19 pandemic. For example, in NYC Citi Bike which is a dock-based bike-share system, there is a nearly 40% increase in the trip duration in March 2020 than months before COVID (Teixeira and Lopes, 2020). Similar increases were also observed in other US cities like Boston and Chicago (Padmanabhan et al., 2021). Another study on NYC Citi Bike concluded the increase in longer trips can suggest the shift from the subway, due to fear of contagion (Wang and Noland, 2021). Such increases in the trip length are not only found in the US, but also seen in some European cities such as London (Heydari et al., 2021), Zurich (Li et al., 2021), and Kosice, Slovakia (Kubalak et al. 2021). Such change is a strong indicator of the replacement of public transit with bike share, but there is no further verification or explanation on this. By looking into different heuristic reasons, this research practice aims to investigate if the link between longer trip duration and replacement of public transit holds or breaks. NYC is selected as the study area as it has the largest usage of bike-sharing and is mostly discussed in the related studies.

Methods

Methodology in achieving the objective of this research contains two parts: First, to confirm the most significant factor causing the longer trip duration, and second, the corresponding explanation. Intuitively, reasons for the trip duration increase might come from the internity of BSS itself, such as people may cycle slower due to the higher traffic speed when sharing roads with fewer motor vehicles during the lockdown, or the system expansion may increase the average distance between an origin station and a destination station. These two hypotheses are rejected by examining the same origin and destination existing in the time period before and after COVID. Under the identical station condition, the hypothesis of unchanged speed cannot be rejected. Yet more trips between longer origin-destination were found since the initial outbreak of COVID.

As the external factor, public transit is usually considered an important factor in the usage of bike-share because a bike-share system is often introduced as complementation or substitution of public transit. Different from public transit which is targeted for long-distance transit, BSS can better address the first- or last-mile problem. From existing studies on the relationship between BSS and public transit, a methodology was proposed to assign mode semantics with public transit for each bike trip (Kong et al., 2020). The bike-share mode semantics might answer the first research question.

To infer the relationship between each bike trip and public transit, each bike-share trip is labeled with mode substitution (MS), mode integration (MI), or mode complementation (MC). We use the MS and MI trips to investigate the replacement of public transit, especially the ridership of NYC Subway, with bike-share trips. Kong et al. first classify stations of trip origin and destination by their accessibility to public transit hubs, including subway stations and bus stops, into three categories. Label A for stations within 100 meters of a subway station or bus stop; label B when less than 400 meters; and label C when outside the 400 meters buffer which is believed to be outside the public transit service area.

Based on the station categories and the trip duration, a rule-based classification tree is deployed to assign each bike trip with one of the three mode semantics: MS, MI and MC. MS represents a relatively long trip that starts and ends close to public transit. MI refers to short trips that start or end close to public transit and represent bike trips that are probably from a combined mode of bike share and public transit. MC include two cases: Long trips or trips with multiple public transit transfers that must start and end close to public transit; second, any trip that starts and ends far from public transit.

Results

The classification system is applied to bike trips in 2019 and 2020. Daily aggregated results of 2019 show that before COVID-19, most trips are modal integration or modal substitution with public transit and less than 10 percent of total trips are complementation of public transit. Winter has fewer MI trips than MS, while during summer times MI and MS have the same share. In 2020 after the COVID-19 started, results show that there was an increase in the mode substitution and a decrease in the mode integration after the outbreak in April 2020. This can naturally result in a longer trip duration. But the increase of MS trips was mostly during weekends which is shown as the red spikes. This means people take many long-distance trips during weekends which is different from 2019 patterns.

Trip semantics in 2019

Trip semantics in 2020

The result provides encouraging evidence that demonstrates a shift from public transit to bike trips. In the second part, the explanation of the substitution is discussed by exploring some heuristic reasons. Intuitively the reasons are 1) the health concern, 2) the extra delay from reduced services, and 3) free membership program. The context of free membership is that during the first two months of COVID-19, free one-year membership of Citi Bike was granted to healthcare workers in NYC. Since the influence of free membership was discussed in the two other papers (Teixeira and Lopes 2020, Pase et al, 2020), the rest two reasons are discussed in the following results. The methods is for delay time analysis is to collect delay information of PT while the travel time matrix of public transit and bike-share is used to evaluate the health concern reasons. Travel time matrix of public transit is pulled from map service TravelTime and the bike-share matrix is computed from the mean value of bike trip duration.

A daily time series plot aligning covid cases, subway delay information, subway ridership number, and bike trip by mode semantics is plotted below. The first plot starting at the top shows the daily count of confirmed COVID cases in NYC. The From delay information, we can see the delay happened following the first wave of covid which last just a month. But the increase of MS trips existed longer than 1 month. The mismatch of timeline downvotes the delay reason. Increasing weekend trips also indicate that the delay factor is not a very significant one.

The travel time matrices of public transit and bike share compare the time trade-off between the same origin and destination. The heatmap shows the difference of trip distribution between two periods, from June to Aug in 2019 and 2020 which is not affected by delays in public transit service. The deep blue blocks below the red reference line indicate bike trips are made even PT has shorter trip time. Besides, bike trip could averagely save 5.5 minutes than taking public transit before COVID but only save 4.6 min during the COVID. Both findings upvote the health concern reason, that people actively chose bike share as a safer mobility option than public transit.

travel time difference in public transit and bike-share

Discussion

To summarize, mode semantics classification and explorative data analysis on various hypotheses help to confirm an increase in the substitution of bike-share over public transit as the reason for longer trip duration in COVID-19. Between the reasons of delay and health concern, this substitution is less possible due to the delay of public transit, as the short period of delay doesn’t overlap the longer substitution period. While it is more possible due to the health concern as 1) the average trip time saving of bike-share decrease and 2) longer bike trip replacing PT still happened after the resume of regular PT service.

Meanwhile, limitations are waiting to be mitigated with a better solution to improve the research methods and results, e.g., Without direct responses from the public, such as surveys, there is no ground truth for the semantics interpretation as well as the reason for the replacement Second, delay analysis is simplified without having a full archival of delay history. Any data sources that can address the limitations are welcome for further exploration.

References