Recent Decline in Public Transportation Ridership: Hypotheses, Methodologies, and Detailed City-by-City Results (2022)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

G-1 Appendix G: Technical Methodology for Impacts of Shared E-Scooters on Bus Ridership This Appendix compliments the analysis results in Chapter 7 of the report. G.1 Data Used in the E-Scooter Analysis This section discusses the data and methodology used for the study. First, the transit data, the shared e-scooters variables, and the other variables used in this analysis are described. Then, the method used to assign shared e-scooter trips to transit routes is discussed. This is followed by the summary statistics of variables used in the analysis. Finally, the modeling framework used is presented. The three different data used in this analysis are described in this section. G.1.1 Transit Data The main dependent variable explored in this study is bus ridership, measured as the weekday unlinked passenger trips per route (UPT). The level of transit service provision per route is one of the major determining factors of transit ridership. Thus, vehicle revenue hours per route (VRH) were used in the study as a measure of transit service provision. These two variables were obtained through a data request from TARC. The geographic locations of bus stops and routes were obtained from a TARC shapefile from the Louisville – Jefferson County open data ArcGIS Hub page (Louisville (LOJIC) Open GeoSpatial Data, 2020). Other changes related to the transit system that could also potentially affect transit ridership, namely the implementation of the MyTARC smart card, were found from TARC press releases (Transit Authority of River City, 2018). G.1.2 Shared E-scooters Trip Data In August 2018, Bird was the first shared e-scooter operator to launch its service in Louisville, with competitor companies Lime, Bolt, and Spin launching shortly thereafter. The initial launch of each of these companies was limited to 150 vehicles per day by the local government in Louisville with the opportunity to increase the fleet size over time to a maximum of 1,050 vehicles per day per operator (Louisville Metro Public Works and Assets, 2019). The Louisville Metro government also put several policies in place to ensure safety and transparency within the community. These policies included restricting the service area for shared e-scooters to primarily downtown Louisville and its surrounding neighborhoods, incorporating no-park and speed limited zones in areas with heavy pedestrian activity, and distributional requirements to ensure no zone was underserved in the service area. Additionally, the Louisville Metro government required all shared e-scooters operators to make anonymized shared e-scooters trip data freely accessible to the public, which was utilized to carry out the analysis performed in this study (Louisville Metro Public Works and Assets, 2019).

G-2 As in many cities nationwide, shared e-scooter usage has grown in Louisville. Figure G-1 presents the average daily ridership of shared e-scooters in Louisville over the period August 2018 through December 2019. Figure G-1 shows that for all months except December 2019, the average number of shared e-scooters daily trips is more than 200 trips per day. Also, it can be noticed that there was a visible increase in the average number of daily trips in the period April 2019 to October 2019; the mild weather is a possible reason for this increase. Shared e-scooter trip data were acquired from the publicly accessible Louisville Metro open data repository (Louisville Metro Open Data, 2020). The shared e-scooter trip file contained anonymized trip data from four shared e-scooters operators (Bird, Lime, Bolt, Spin) from August 2018 through December 2019. Each shared e-scooter trip contains a unique identifying number, trip start date, end date, time, and the latitude and longitude for the trip origin and destination. In addition, the Louisville Metro open data website has a shapefile for the designated shared e- scooter service area as shown in Figure G-2. Figure G-2 shows the shared e-scooters service area and TARC transit routes. Figure G-1: Average daily shared e-scooters trips for Louisville for the period August 2018 to December 2019

G-3 Figure G-2: Shared e-scooters service area and TARC transit routes The shared e-scooter trip data were cleaned by implementing several filters related to trip viability. Shared e-scooter trips were deemed viable if their origin and destination both fell within the shared e-scooter service area and if the trips had a non-negative, non-zero trip duration or distance. Additionally, only trips of less than 80 minutes in duration and between 250 feet and 10 miles in distance were considered. These limits were selected to identify real trips and not data anomalies. Finally, shared e-scooter trips were limited to those with an average speed less than or equal to 15 mph (the maximum speed limit for shared e-scooters in Louisville), as an average speed higher than the maximum could be attributed to the shared e-scooter being activated on an alternate form of transport for rebalancing or other purposes. Of the approximately 501,000 original trips, 113,329 trips were removed due to these constraints, leaving 383,500 shared e-scooter trips. This represents 76.5% of the original shared e-scooter trips. In the cleaned data set, the average trip duration was about 14 minutes, and the average trip distance was 1.20 miles. More than 50% of shared scooter trips occurred between the hours of 12:00 and 18:00. Also, about 32% of shared e-scooters trips were taken on Saturday and Sunday, indicating a somewhat higher demand for shared e-scooters during weekends, especially compared to typical transit ridership patterns. The cleaned shared e-scooter trip data set and transit route shapefiles were then used to assign the shared e-scooter trips to the different transit routes, the process of which is described in Section 5.2.

G-4 G.1.3 Other Variables This study also used other variables that may affect transit ridership like population, income, employment, gas price, and weather. The one year American Community Survey (ACS) estimates were used to provide the population and annual median income of individuals for the Louisville Metropolitan Statistical Area (MSA) over the period of analysis. Employment data for the Louisville MSA was obtained from the Bureau of Labor Statistics, and gas prices were obtained from the U.S. Energy Information Administration website. Weather data that include the average temperature, rain fall, and snow fall were obtained from National Oceanic and Atmospheric Administration. It also worth noting that bikeshare system in Louisville (LouVelo) started operation in 2017 (Finley, 2017), which will be considered in the modeling approach as discussed in Section 5.4. G.2 Methodology G.2.1 Assigning Shared E-scooter Trips to Transit Routes This section discusses the method used to assign shared e-scooter trips to transit routes for both local and express routes. To assign shared e-scooter trips to transit routes, first, a transit catchment area of 0.1 mile was defined around each bus route. The transit catchment area was defined as 0.1 mile for two reasons: first, shared e-scooters are dockless, and users could park them near their final destination, which, in this case, would be the bus stop. Second, the 0.1 mile radius of the catchment area reduced the number of trip that were assigned to more than one bus route due to overlapping bus route catchment areas in the downtown area. Then, using this catchment area and shared e-scooter trip origins and destinations, different variables were defined for each bus route using the Geopandas library in Python, as discussed in the remainder of this section. G.2.1.1 Shared E-scooter Trips Assignment to Local Bus Routes The first variable defined for local routes was the shared e-scooter substitute trip count. This variable was defined to explore the first research question of this study: do shared e-scooters decrease ridership on local bus routes? Local bus routes were selected as the primary transit routes that could be impacted by competitive shared e-scooter usage as the bus stops are located every 0.1 to 0.25 miles throughout downtown Louisville. This is also the predominant location for shared e-scooter trips and the majority of the shared e-scooter trips analyzed overlap with the bus routes. A shared e-scooter trip was considered a potential substitute for a transit trip if the trip distance was greater than 0.1 miles and both the shared e-scooter trip origin and destination points were located within the catchment area for a specific local bus route. The total count of potential substitutionary shared e-scooter trips was determined for each local bus route for each day of the period of analysis. Figure G-3 presents an example of the shared e-scooter trip assignment method used for one day of trip data (August 10, 2018) for one local bus route shown as the blue line (TARC Route 23

G-5 Broadway). In Figure G-3, the orange line shows the catchment area for Route 23. The shared e- scooter trip origins are shown as red dots, and the destinations are shown as green dots in the figure. The full extent of the map shows the cleaned shared e-scooter trip origins and destinations in the service area, as well as the transit route catchment area. The inset map shows the trips that would be counted for this day and route: the origin and destination for these 34 scooter trips (which are shown as “linked” in the inset) are both within the catchment area for the transit route, implying that these scooter trips could have replaced transit trips along this path. Figure G-3 Example of shared e-scooter trips assignment to a local TARC transit route (Route 23 Broadway) for August 10, 2018 Figure G-4 summarizes the results of assigning the shared e-scooter trips to local bus routes in the form of a heat map. The different colors in Figure G-4 show the variation in the count of possible shared e-scooters trips along each route. Notably, the blue lines show routes that have no shared e-shared scooter trips within their catchment area. These routes serve as control group in the regression analysis. The second research question of this study assesses the possibility that shared e-scooters increase ridership on local bus routes since they could serve as first/last-mile connectors. In a complementary relationship between shared e-scooters and local bus routes, a rider could use a shared e-scooter to get to/from bus stops. Therefore, two additional variables were defined for

G-6 local routes to assess the second research question. The variable shared e-scooter first-mile connector trip count was defined as the number of shared e-scooter trips that have destinations within the bus route catchment area. The hypothesis for this variable is that users could ride a shared e-scooter to get to a local bus station. This variable is used to evaluate the possibility that shared e-scooters are used as first-mile connectors from the trip starting point to a local bus stop. The second variable is the shared e-scooters last-mile connector trip count. This variable counts the number of shared e-scooter trips that have origins within a bus route’s catchment area. The assumption is riders could take a shared e-scooter from the bus stop to their final trip destination. This variable was used to evaluate the possibility that shared e-scooters are used as last-mile connectors from a local bus stop to the user’s trip end point. Figure G-4: Heatmap of average shared e-scooters substitute trip count for each local TARC route for the period August 2018 to December 2019

G-7 G.2.1.2 Shared E-scooter Trips Assignment to Express Bus Routes Express bus routes serve longer bus trips that have one of their ends located outside the shared e- scooters service area in Louisville. Therefore, it is unlikely that express bus trips could be replaced by a shared e-scooter. Instead, shared e-scooters could potentially complement express routes as they provide first/last mile access to transit. Therefore, the shared e-scooters complement trip count was defined for express routes only to assess the third research question that shared e-scooters increase ridership on express bus routes. A shared e-scooter trip was defined as complementary for an express bus route if the shared e- scooter trip origin was located within an express route bus stop catchment area during morning hours, or if the shared e-scooter trip destination was located within the catchment area during evening hours. The assumption is that a commuter exits the express bus in downtown in the morning and then could begin a shared e-scooter trip for the last portion of their commute. In the afternoon, a commuter could ride a shared e-scooter from work to the express bus stop for the first portion of their evening commute, and then board the express bus for their trip home. Morning trips correspond to shared e-scooter trips whose start times were between 4:00 and 10:00, while evening trips correspond to shared e-scooter trips whose end times were between 13:00 and 20:00. The morning and evening hours were selected as they correspond to the hours of service for the TARC express bus routes. It should be noted that since express routes do not serve every stop along their path, the 0.1-mile catchment area was created only for the stops that are served by the express route and fall within the shared e-scooter service area. Figure G-5 presents an example of the shared e-scooter trip assignment method used for one day of trip data (August 10, 2018) for one express bus route (TARC Route 65 Sellersburg Express). The larger map shows the shared e-scooter trip data that falls within the express service time restrictions, as well as the path of the express route. The inset map shows express route stops with their catchment areas, as well as the shared e-scooter trip origins and destinations that fall within them. Six shared e-scooter trips originated within the catchment areas during morning express service hours and 26 shared e-scooter trips ended within the catchment areas during afternoon/evening express service hours, implying that shared e-scooter trips could potentially complement up to 32 transit trips on this express route on this day. Figure G-6 shows a heat map of the average shared e-scooters complement trip count for TARC express routes. The average shared e-scooters complement trip count ranged between 40 and 79 trips per day. TARC express routes end with a similar loop in downtown Louisville, and many of the express routes serve the same stops, so the range was similar for most routes. It should be noted that the fact that many express routes serve the same stops is one of the limitations of this research design since the variability between shared e-scooters complement trip count for different express routes is limited.

G-8 Figure G-5: Example of shared e-scooter trips assignment to an express TARC transit route (Route 65 Sellersburg Express) for August 10, 2018

G-9 Figure G-6: Heatmap of the average daily shared e-scooters complement trip count for each express TARC route for the period August 2018 to December 2019 G.2.2 Summary Statistics Table G-1 presents the summary statistics of the different variables. First, the average bus ridership was 899 unlinked passenger trips per day per route with ridership ranging from 1 to 9,632 unlinked passenger trips per day per route. Second, the average shared e-scooters substitute trip count was 44 shared e-scooter trips per day per route, which represents 5% of the average daily bus route ridership. Third, the average shared e-scooters complement trip count was 68 shared e-scooter trips per day per route, which represents about 7.5% of the average daily bus route ridership. The shared e-scooter first-mile connector trip count and shared e-scooter last-mile connector trip count had similar summary statistics. This is likely because shared e-scooters trips in Louisville are typically short trips and most of them occur downtown.

G-10 Table G-1: Summary Statistics Variable Spatial Unit Temporal Unit Min Max Median Mean (Standard Deviation) Dependent variable Unlinked passenger trips a Route Daily 1 9,632 293 899 (1,451) Transit variables Bus vehicle revenue hours (VRH) 2.5 241 23 40 (60) Shared e- scooters variables Shared e-scooters substitute trip count (Local routes) 0 607 21 44 (55) Shared e-scooters complement trip count (Express routes) 0 276 54 68 (50) Shared e-scooter first-mile connector trip count (local routes) 0 877 52 91 (107) Shared e-scooter last-mile connector trip count (local routes) 0 894 53 94 (113) Other variables Population + employment b (in 1000s) City Monthly 1897.99 1970.60 1946.23 1941.15 (20.04) Annual median income of individual ($) Monthly 29,509 34,833 32,134 32,132 (1553.31) Average temperature (ºF) Daily 9.00 89 62 60 (18.05) Rain fall (inch) 0.00 3.84 0.00 0.14 (0.38) Snow fall (inch) 0.00 4.90 0.00 0.03 (0.27) a Trips for a few routes were merged due to combined ridership counts. b Monthly population estimates were generated using linear interpolation.

G-11 G.2.3 Modeling Framework This section discusses the modeling approach used in this analysis. This study evaluated the research questions by estimating several fixed effects regression models of the change in route- level bus ridership as a function of the change in shared e-scooters trips within a catchment area near each route. The study period includes both the period before and after the shared e-scooters were introduced and controls for other important determinants of transit ridership. This basic approach has been successfully applied in the past to study the effect of traveler information on transit ridership and other determinants of changing transit ridership (Brakewood et al., 2015; Graehler et al., 2019). The fixed effects regression equation is shown in Equation 1 (Studenmund, 2016). The dependent variable is the bus unlinked passenger trips per route per day. The explanatory variables are vehicle revenue hours, shared e-scooter trip counts, and other external control variables like population and employment, income, and weather. 𝑦𝑦𝑖𝑖𝑖𝑖 = 𝑥𝑥𝑖𝑖𝑖𝑖 ∗ β + 𝛼𝛼𝑖𝑖𝐸𝐸𝐹𝐹𝑖𝑖 + 𝜌𝜌𝑖𝑖𝑇𝑇𝐹𝐹𝑖𝑖 + 𝜀𝜀𝑖𝑖𝑖𝑖 (1) Where: • yit: unlinked passenger trips for bus route i during time t (day) • xit: explanatory variables for bus route i during time t (e.g., shared e-scooter trip counts, vehicle revenue hours, population and employment) • 𝐸𝐸𝐹𝐹𝑖𝑖: entity fixed effect dummy, equal 1 for bus route i and 0 otherwise • 𝑇𝑇𝐹𝐹𝑖𝑖: time fixed effect dummy, equal 1 for the tth period and 0 otherwise • εit: error term This approach lets us model the change in transit ridership as a function of the change in explanatory variables. This focus on the change is important because we know that both transit ridership and shared e-scooter ridership are concentrated in the same areas—downtown and surrounding urban neighborhoods. This means that any cross-sectional regression model would show a high correlation between the two, but that correlation may be driven by the same favorable conditions rather than complementarity. The fixed effect term captures unobserved variables at the route level (such as serving a transit-favorable area) and more directly measure whether routes with a high number of competitive or complementary shared e-scooter trips change ridership at a different rate than other routes. In addition to bus route fixed effects, this study also uses time fixed effects, which adds a dummy variable for each time-period but one (Studenmund, 2016). The time fixed effects control for any unobservable variable at a specific date like special events as well as other systemwide changes like the launch of MyTARC app and the bike share system (Studenmund, 2016).

G-12 This study also used the clustered sandwich estimator to estimate cluster-robust standard errors (StataCorp LLC, 2017). The cluster-robust standard errors are robust to heteroskedasticity and serial correlation (StataCorp LLC, 2017; Wooldridge, 2012). This analysis was perform using Stata “xtreg” command. G.3 Results This section discusses the results of this study. This section is divided into three parts. The first part of the results seeks to answer the first research question: do shared e-scooters decrease ridership on local bus routes? The second part assesses the second research question: do shared e-scooters increase ridership on local bus routes? It should be noted that the third research question is discussed in both parts since the relationship between shared e-scooters and express routes could only be complementary. The final part discusses these results and their implications for practice. G.3.1 Research Questions 1 & 3: Do Shared E-Scooters Decrease Ridership on Local Bus Routes and Increase Ridership on Express Bus Routes? Table G-2 shows the results of evaluating the first research question, “Do shared e-scooters decrease ridership on local bus routes?” and the third research question “Do shared e-scooters increase ridership on express bus routes?” To answer these questions, four different model specifications were estimated, as shown in Table G-2. Models 1,2, and 3 used the same explanatory variables to estimate the weekday, weekly, and the monthly models, respectively. Model 4 is an expanded version of the monthly model (Model 3), which includes additional explanatory variables such as population and employment. Model 1 in Table G-2 shows the weekday model, which used the daily unlinked passenger trips for Tuesday to Friday as the dependent variable and the VRH, the shared e-scooter substitute trip count (local routes), and the shared e-scooter complement trip count (express routes) as explanatory variables. The reason Monday data were excluded from this model is because TARC archives Monday and weekend ridership together. Model 1 also included route fixed effects and the day fixed effects. The route and the day fixed effects control other unobservable variables not included in the model. In Table G-2, Model 1 shows VRH is a highly significant predictor for bus ridership, as indicated by the significant positive coefficient (33.3). Model 1 also indicates that the shared e-scooter substitute trip count for local bus routes is not a significant predictor for bus ridership. This suggests that shared e-scooters are not decreasing ridership on local bus routes in Louisville. On the other hand, Model 1 suggests that the shared e-scooters complement trip count for express bus routes is a significant predictor for bus ridership, as indicated by the significant positive coefficient (0.61). This coefficient indicates that every 10 shared e-scooter trips occurring within the catchment area of an express bus route stop could potentially complement up to 6 express bus trips, holding all else constant. However, this result should be interpreted with caution as discussed later in this section.

G-13 Model 2 shows the estimated weekly model with the same specification as the weekday model. This model was estimated since there is a high portion of shared e-scooter trips occurring on Saturdays and Sundays, which might have impacts on bus ridership. However, the results of Model 2 are very similar to Model 1. Model 3 shows a monthly model with a similar specification as Models 1 and 2. The motivation for this model is daily and weekly ridership might be impacted by many different variables; therefore, a monthly model could yield more robust results. The results of Model 3 are also similar to Models 1 and 2 with a slightly higher estimation of the shared e-scooters complement trip count coefficient. Model 4 in Table G-2 includes additional variables that affect transit ridership such as population, employment, income, and weather. The results of Model 4 are consistent with Models 1-3; the shared e-scooters complement trip count is a significant predictor for express bus ridership. Model 4 also shows that population and employment have a positive impact on ridership, but it is not significant. The reason behind this is likely the small change in population and employment during the analysis period. Model 4 also shows that income, rainfall, and snow have significant negative impacts on bus ridership, which aligns with findings from prior studies (Brakewood et al., 2015; Ngo, 2019; Owen and Levinson, 2015). Table G-2: Fixed Effects Regression Results (Models 1–4) Dependent variable: unlinked passenger trips per route (1) Weekday Model (2) Weekly Model (3) Monthly Model (4) Expanded Monthly Model VRH 33.30*** (8.34) 28.90*** (2.88) 31.00*** (3.03) 31.00*** (3.03) Shared e-scooters substitute trip count (Local routes) -0.07 (0.43) -0.06 (0.31) -0.19 (0.30) -0.05 (0.29) Shared e-scooters complement trip count (Express routes) 0.61** (0.27) 0.61** (0.24) 0.86*** (0.29) 0.66** (0.27) Population and employment (in 1000s) 6012.00 (4992.00) Annual median income of individual ($) -80.10* (44.70) Average temperature (ºF) -292.00 (219.00) Rain fall (inch) -898.00* (460.00) Snow fall (inch) -1281.00*** (340.00) Route fixed effect Yes Yes Yes Yes Time Fixed Effect Day Week Month Month R2 0.256 0.576 0.615 0.625 N 33081 8521 1980 1980 Clustered robust standard errors are shown in parenthesis. *p<0.10; **p<0.05; ***p<0.01

G-14 The findings of Models 1-4 suggest that shared e-scooters have no significant effect on local bus ridership. They also suggest that e-scooters may increase ridership on express bus routes in Louisville. However, this result should be interpreted with caution. In Louisville, all express routes terminate in the same geographic area near downtown, which resulted in similar counts of shared e-scooter trip counts along all express routes (Figure G-6). These similar counts limit the variability of this variable in the model, which is one of the limitations of this experimental design. Therefore, the relationship between express bus ridership and shared e-scooters requires further study. G.3.2 Research Questions 2 & 3: Do Shared E-scooters Increase Ridership on both Local and Express Bus Routes? Table G-3 shows the results for the second and the third research questions – that shared e- scooters increase ridership on both local and express bus routes. To answer these questions, different model specifications were estimated, as shown in Table G-3. Models 5 and 6 present the results of the weekday models. Models 7 and 8 show the weekly models, followed by Models 9 and 10, which present the results of the monthly models. The expanded monthly models are presented by Models 11 and 12. Model 5 evaluates the hypothesis that shared e-scooters may increase ridership on local bus routes as they serve as last-mile connectors, while Model 6 evaluates the hypothesis that shared e-scooters increases ridership on local bus routes as they serve as first-mile connectors. The results of these models show that both the shared e-scooter last-mile connector trip count and the shared e-scooter first-mile trip connector count are not significant predictors for bus ridership in Louisville, as indicated by the insignificant coefficients (-0.11) and (-0.12) in Models 5 and 6, respectively. This finding suggests that shared e-scooters are not increasing ridership on local bus routes in Louisville. Also, the results of both Models 5 and 6 indicate that bus VRH is a highly significant predictor for bus ridership, as indicated by the significant positive coefficient (33.3). Furthermore, both models show that the shared e-scooter complement trip count is a significant predictor of bus ridership as indicated by the significant positive coefficient (0.55). This finding is consistent with Models 1-4, the findings from which are discussed in Section 6.1. Models 7-12 use weekly and monthly bus ridership and shared e-scooter trip counts to evaluate the same research questions. The results of Models 7-12 are consistent with findings of Models 5 and 6 that shared e-scooters are not complementing local bus ridership as first/last-mile connections. Models 11 and 12 suggest that population and employment have a positive association with bus ridership, but that it is not significant. Models 11 and 12 also show that income, rainfall, and snow have significant negative impacts on bus ridership, which aligns with finding from Model 4 and prior studies (Brakewood et al., 2015; Ngo, 2019; Owen and Levinson, 2015).

G-15 Table G-3: Fixed Effects Regression Results (Models 5–12) Dependent variable: unlinked passenger trips per route (5) Weekday Model (6) Weekday Model (7) Weekly Model (8) Weekly Model (9) Monthly Model (10) Monthly Model (11) Expanded Monthly Model (12) Expanded Monthly Model VRH 33.30*** (8.36) 33.30*** (8.33) 28.80*** (2.88) 28.80*** (2.88) 31.00*** (3.02) 31.00*** (3.03) 31.00*** (3.02) 30.90*** (3.1) Shared e-scooter last- mile connector trip count (local routes) -0.11 (0.22) -0.06 (0.17) -0.13 (0.16) -0.13 (0.16) Shared e-scooter first- mile connector trip count (local routes) -0.12 (0.22) -0.06 (0.18) -0.14 (0.17) -0.14 (0.17) Scooter complement trip count (express bus) 0.55** (0.26) 0.55** (0.26) 0.68** (0.27) 0.68** (0.27) 0.83*** (0.29) 0.82** (0.29) 0.65*** (0.29) 0.65** (0.27) Population and employment (in 1000s) 6004.00 (4998.00) 6012.00 (4992.00) Annual median income of individual ($) -79.80** (44.5) -80.10** (44.7) Average temperature (ºF) -291.00 (219.00) -292.00 (219.00) Rain fall (inch) -894.00* (456.00) -898.00*** (460.00) Snow fall (inch) -1282.00*** (341.00) -1281.00*** (340.00) Route fixed effect Yes Yes Yes Yes Yes Yes Yes Yes Time fixed effect Day Day Week Week Month Month Month Month R2 0.255 0.255 .576 .576 0.616 0.616 0.625 0.625 N 33081 33081 8521 8521 1980 1980 1980 1980 Clustered robust standard errors are shown in parenthesis. *p<0.10; **p<0.05; ***p<0.01 G.4 Discussion and Implications for Practice This study performed an empirical analysis to quantify the impacts of the shared e-scooters on citywide bus ridership. The results of this study suggest that shared e-scooters do not have a significant impact on local bus ridership in Louisville. This finding could be explained by two factors. First, transit and shared e-scooters are typically used for different purposes. Transit in Louisville is mainly used for work and school; a recent survey indicates that 70% of TARC trips are for work and school (Copic, 2019). However, prior studies of shared e-scooter trip patterns indicated that shared e-scooters might not be used for commuting and are likely used for recreation (Caspi et al., 2020; Noland, 2019). Furthermore, the difference in trip length between these modes also suggests these two modes are being used for different purposes. The average transit trip length in Louisville is 4.2 mile compared to 1.2 mile for shared e-scooters (NTD Transit Agency Profiles 2020). These different trip characteristics (both purpose and length) suggest that shared e-scooters are not used to replace transit trips in Louisville. Second, transit users in Louisville are typically minorities and have lower household income (Copic, 2019),

G-16 which is likely different from typical shared e-scooter users, who are likely to be white and have higher household income, as suggested by surveys from other cities (Mobility Lab and Arlington County Commuter Services (ACCS) 2019; San Francisco Municipal Transportation Agency, 2019). These demographics suggest that transit and shared e-scooters are likely used by different groups of riders, which might limit the interaction of these two modes. This said, exploring the relationship between shared e-scooters and express bus routes in Louisville does suggest that shared e-scooters could increase ridership on express bus routes as first/last-mile connectors. However, as discussed in Section 6.1 this finding should be interpreted with caution as the relationship between express bus ridership and shared e-scooters requires further study. However, transit agencies and shared e-scooters operators could explore ways to integrate these two modes to offer better service for their users. There are many ways that such an integration could occur, such as developing multimodal trip planning platforms and price bundling. G.5 References Brakewood, C., Macfarlane, G.S., Watkins, K., 2015. The Impact of Real-Time Information on Bus Ridership in New York City. 53, 59-75. Caspi, O., Smart, M.J., Noland, R.B., 2020. Spatial associations of dockless shared e-scooter usage. Transportation Research Part D: Transport and Environment 86, 102396. Copic, A., 2019. Equity in Action, APTA Mobility Conference, Louisville, KY. Finley, M., 2017. Louisville bike-share program, LouVelo, to launch this spring, https://www.bizjournals.com/louisville/news/2017/04/14/louisville-bike-share-program-louvelo- to-launch.html, Sep 18, 2020 Github, 2020. Mobility Data Specification, https://github.com/openmobilityfoundation/mobility- data-specification, Oct 19,2020 Graehler, M., Mucci, R.A., Erhardt, G.D., 2019. Understanding the Recent Transit Ridership Decline in Major US cities: Service Cuts or Emerging Modes. Presented at 98th Annual Meeting of the Transportation Research Board, Washington, D.C. Louisville (LOJIC) Open GeoSpatial Data, 2020. Louisville Metro Area KY TARC Bus Routes, https://data.lojic.org/datasets/louisville-metro-area-ky-tarc-bus-routes-/data, Mar 26, 2020 Louisville Metro Open Data, 2020. Dockless Vehicle Trips - Block Level - 2018-08 to 2020-01, https://data.louisvilleky.gov/dataset/dockless-vehicles/resource/e36546f6-888b-4e66-8a87- 9b68cab471e6#{}, March 26, 2020 Louisville Metro Public Works and Assets, 2019. Louisville Metro Public Works & Assets Dockless Vehicle Policy, Louisville.