In addition to choosing attributes based on significance, we considered the importance of attributes for policy implications. For this reason, we kept the densities of bike lanes and bike racks even though they were not statistically significant in the final model. The major limitations of this study stem from the nature of the bike sharaing activity data that is used. The time period observed is the first full month of JUMP operations in San Francisco, which is likely to include travel behavior of early adopters and novelty rides that do not reflect more regular patterns that may have emerged among JUMP users since its launch. In addition, by comparing JUMP and GoBike trips during this time period, we observe the interdependent effects of both the dockless model and the electric pedal-assist bicycles on JUMP travel behavior compared to that of GoBike, which used non-electric bicycles with a stationbased model during the study period. While differences in travel behavior related to elevation may be more directly linked to the e-bikes in the JUMP system, most other trip attributes examined may be influenced by a number of variants in the operation and/or ridership across the two bike sharing systems. The lack of user data linked to the trips we observe constrains our ability to account for socio-demographic differences across the riders of the two systems. We use census data to differentiate bike sharing trip destinations by the socio-demographic makeup of the surrounding census tracts, though we cannot directly draw conclusions about the sociodemographic characteristics of riders, nor of the actual points of interest visited during each trip. In addition, we used suggested bike routes from the Google Directions API to estimate trip distances, durations, and elevation gain in the absence of trajectory data. However,trimming tray we chose not to incorporate bike path availability along these suggested routes in the DCA model due to a concern that the results would overestimate the use of bike routes. Lastly, there is a degree of endogeneity in our DCA results for GoBike, as the destination choices of GoBike users are completely constrained to the station locations of the GoBike system.
We begin with a visual analysis of the geographical and temporal distribution of demand for each bike sharing system. Figures 4.a – d. display heat maps of bike sharing activity during February 2018 by time of day. Areas in which the departures constitute the majority of activity are shaded green, while areas in which arrivals constitute the majority of activity are shaded red. Thirty-two percent of JUMP trips, 33% of GoBike non-member trips, and 43% of GoBike member trips took place during the AM period . Both JUMP and GoBike exhibit concentrated AM demand destined for dense employment centers along Market Street and in the South of Market and Financial District areas just South-East and North-West of Market Street, respectively. These neighborhoods are home to many large office buildings housing numerous corporate headquarters and branch offices. The intensity of trip arrivals around the Civic Center could represent multi-modal trips, as bike sharing users may choose to transfer to the Bay Area Rapid Transit line at this most North-Western access point. There is a clear difference in the trip origins of JUMP and GoBike in the AM period, where we see a concentration of GoBike trips departing from the CalTrain and Embarcadero BART stations, while JUMP trip departures were spread out in neighborhoods outside of the CBD. In the PM period , we observed both systems servicing riders originating in the CBD, but the destinations of JUMP trips were again spread out in neighborhoods farther away from the CBD, while GoBike trip destinations were concentrated at the Caltrain and Embarcadero BART stations.The distribution of bike sharing trip distance and duration for each of the two systems exemplifies the behavior observed in the visual analysis. We assess the trip characteristics of GoBike members and non-members separately, noting that 95% of GoBike trips were made by members. JUMP trips tended to be longer in distance and duration than GoBike trips . The average JUMP trip was about a third longer in distance and about twice as long in duration as the average GoBike member trip. While this may be a result of the newness of JUMP in February 2018, the similarity implies that JUMP tended to be used for longer, potentially more recreational trips, which are more similar to GoBike non-member trips.
Indeed, 7% and 8% of JUMP and GoBike non-member trips, respectively, are longer than one hour in duration, compared to less than a third of a percent of GoBike member trips. Unlike GoBike members who pay on an annual basis and are incentivized to make the most out of their membership regardless of trip length, JUMP users pay per trip and thus may prefer to make longer, less frequent trips. Next, we present the results of the destination choice model estimation to better understand the influence of different factors on bike sharing users’ destination choices in the GoBike and JUMP systems . The final log-likelihood values for the destination choice models for GoBike and JUMP trips were -1,402 and -1,713, respectively. The R squared values for the models were 0.26 for GoBike and 0.27 for JUMP, while the R bar squared values for each were 0.23 and 0.24, respectively. Across both systems, increase in estimated trip distance and elevation gain were both strong negative factors in bike sharing users’ destination choices. The estimated total elevation gain was by far the most negative coefficient in the GoBike model, indicating that destinations that involved climbing in elevation were very unpreferable to GoBike users. The coefficient for estimated distance in the JUMP model was more negative than that of estimated total elevation gain. The range of estimated trip distances for the destination choice set was inherently larger for JUMP than for GoBike, since JUMP users were not entirely restricted by the service area of the system. Seven percent of JUMP trips in our dataset were completed outside of the service area. JUMP users were fined for ending a trip outside of the service area, but they were not prohibited from doing so. The large positive coefficient for the JUMP service area indicator reflected this incentive. Factors indicating the level of activity at a destination were significant and positive across models. In addition, the density of the resident population and the ASC for low-density residential census tracts were both significant, negative coefficients in the JUMP model, suggesting an affinity of JUMP users to travel to lower-density destinations. Conversely, the model results support our findings from visual spatial analysis that GoBike users were largely bike sharing to work, as the activity level parameters were the two most positive coefficients, and the employment center ASC was positive and significant.
The age and income characteristics of destinations were mostly insignificant in the model. JUMP destination choices were significantly negatively influenced by the fraction of residents over the age of 55 in a destination census tract. The median income of destinations is not a significant factor in the destination choice models for either system, with coefficient estimates close to zero in both models. While bike rack density is unsurprisingly an insignificant factor in the GoBike model, it is a significant positive factor in the destination choices of JUMP users. Since JUMP users were instructed to lock the bikes to public racks,weed trimming tray this finding has two possible implications for the destination choices of dockless bike sharing users: 1) JUMP users may prefer destinations with a higher availability of public bike racks with which to easily end their trips, and/or 2) the spatial distribution of public bike racks is well suited to the preferred destinations of dockless bike sharing users. On a related note, the density of GoBike stations in a destination tract was a significant positive factor in the GoBike model. Again, the location of docking stations may attract users, and/or they are well-placed to serve the destination preferences of GoBike users. The insignificance of bike lane density in both destination choice models may be an artifact of our choice to model this factor as an alternative attribute rather than a trip attribute. Bike lane density along suggested destination routes or even the cumulative bike lane density across each of the census tracts along the destination route may provide a more explanatory variable with which to assess the bike sharing user sensitivity to bike lane availability. The composite suitability maps reveal the geospatial distribution of the “bike ability” for users of JUMP and GoBike in San Francisco. In particular, residential neighborhoods in the Northwest and along the Northeast of the city provide opportunity for expansion for both systems to improve equity based on physiological and economic factors. The distribution of the population over 55 and elevation in these neighborhoods appear to be the main constraints in these areas, while considerable job density and available bike facilities provide opportunities. Though e-bike sharing has potential to overcome physiological barriers for older residents in these areas, considerable social barriers may exist since JUMP is only accessible through a smartphone application. Additional social constraints, which are not visualized in the bike ability maps, may stem from language barriers or broader cultural differences across the city. Finally, introduction of temporal variables would aid in assessing the opportunities and challenges for equitable bike sharing based on the time of day. Bike ability may vary across time periods with different levels of congestion, or across hours of daylight versus darkness. Shared micromobility service models are growing across the U.S. including: docked, dockless, and e-bike sharing models.
Our research analyzes the trip making behavior of JUMP dockless ebike sharing and GoBike docked bike sharing users in the first month of the JUMP pilot program. Travel behavior and destination choice analyses reveal that the two systems appear to complement one another: GoBike trips tended to be short, flat commute trips, mostly connecting to/from major public transit transfer stations while JUMP trips were longer, more spatially distributed and more heavily servicing lower-density neighborhoods. The average JUMP trip was about a third longer in distance and about twice as long in duration than the average GoBike trip. In addition, JUMP trips underwent about three times the elevation gain per trip, on average, compared to GoBike trips. Our findings suggest that the assistance provided by e-bikes in addition to the flexibility afforded by the dockless model are serving mobility demand outside the dense urban core of the city, where docked models are not available. Furthermore, we found that the destination choices of docked bike sharing users are positively influenced by the density of stations, and bike rack density was a significant positive factor for JUMP users. The location of facilities necessary to use either the docked or dockless system may attract users and/or be well-placed to the destination preferences of users. While the sensitivity of destination choices to factors influencing equity, such as older age are slight, our bike ability analysis reveals that the composite effect of constraints and opportunities that impact bike sharing demand can have adverse effects in neighborhoods otherwise ripe for bike sharing expansion. Additional research is needed to more closely link the characteristics of shared micromobility users with differences in travel behavior across business models and service areas. This study focuses on San Francisco, a city with unique topographic, sociodemographic, and cultural features which have distinct effects on travel behavior that may not be generalizable to other locations. As policies and guidelines for shared micromobility are being piloted and refined, similar data sources to those used in this study complemented with user surveys can be used to monitor the emerging trends in ridership across multiple shared modes. Research into the multi-modal trip making and trip chaining using shared micromobility is needed to further the understanding of the potential positive impacts of electric and dockless models on overall mobility and accessibility across trip purposes. Finally, time series analysis of travel behavior before, during, and after the implementation of innovative policies would provide invaluable insights to help hone public interventions strategies that effectively bolster mobility while promoting sustainability and equity within the broader transportation system. The unexpected outbreak of e-cigarette or vaping-associated lung injury was reported nationwide beginning in September 2019, causing more than 2800 hospitalizations and 60 deaths.The specific biological mechanisms of EVALI, as well as the chemical causes, are still under investigation.Evidence shows that EVALI is associated with vaping tetrahydrocannabinol containing e-liquid cartridges that were obtained on the black market.