Predicting Where Transit Ridership Is Under- or Over-Performing Across the SEPTA Network

Author

Sujanesh Kakumanu, Matthew Levy, Kavana Raju

Published

December 8, 2025

Executive Summary

This analysis identifies SEPTA bus stops that are under- or over-performing relative to their predicted ridership based on service characteristics, demographics, and accessibility. By mapping the residuals from our spatial predictive models, we reveal neighborhoods with unmet transit demand and underutilized services, providing actionable recommendations for SEPTA bus service planning and investment priorities.

Key Findings:

1. Introduction

Research Question: Where should SEPTA prioritize investments to maximize ridership and serve unmet transit demand?

Transit agencies operate under persistent budget, fleet, and staffing constraints. As a result, they must make deliberate choices about where to add service, upgrade stops, or scale back underperforming routes. Traditional performance metrics such as raw ridership counts or simple boardings-per-hour thresholds identify busy or quiet locations, but they do not answer a critical question: Is a stop performing better or worse than expected given its context?

A low-ridership stop in a low-density, auto-oriented neighborhood may be performing as expected, while a similarly low-ridership stop in a dense, transit-dependent corridor could signal unmet demand or barriers to access.

To move beyond raw counts, we use predictive modeling to estimate “expected” ridership at each stop based on service characteristics, neighborhood demographics, and accessibility. We then compare these predictions to actual ridership. Stops that overperform (higher ridership than predicted) and those that underperform (lower ridership than predicted) reveal where the network is over- or under-delivering relative to its potential.

This residuals-based approach allows us to:

  • Distinguish between structurally low-demand locations and places where transit is underperforming.

  • Identify corridors with latent demand that could benefit from targeted investments.

  • Highlight stops where ridership is strong despite barriers, suggesting supportive local conditions that can be replicated elsewhere.

Why This Matters

For SEPTA, this analysis supports:

  • Data-driven identification of underserved areas – We flag stops and neighborhoods where ridership is lower than expected despite favorable conditions (e.g., frequent service, high density, transit-dependent populations).
  • More strategic capital and service planning – Residuals provide an evidence-based way to prioritize investments such as stop amenities, service upgrades, or bus-priority treatments.
  • Insight into barriers to ridership growth – Underperforming stops may indicate obstacles such as safety concerns, poor walk access, or reliability issues not directly captured in the data.
  • An equity lens on service allocation – By intersecting residuals with demographic indicators, we can see whether disadvantaged communities are systematically underserved or where high-performing service is already supporting transit-dependent riders.

The remainder of the project builds a modeling and spatial analysis framework to quantify these patterns and translate them into planning guidance for SEPTA’s bus network.

2. Data Sources and Integration

We integrate four main categories of data: 1. Stop-level ridership 2. Transit service characteristics 3. Neighborhood socio-demographics 4. Accessibility and spatial context All spatial data are ultimately projected to EPSG:2272 (PA State Plane South) to ensure consistent distance and area calculations in the Philadelphia region.

Setup

Show code
# Set options
options(scipen = 999)
options(tigris_use_cache = TRUE)

# Packages
if(!require(pacman)){install.packages("pacman"); library(pacman, quietly = T)}
p_load(tidyverse, lubridate, sf, tigris, tidycensus, riem, viridis, gridExtra, knitr, kableExtra, tidytransit, stringr, spdep, corrplot, car, MASS, conflicted, caret, lmtest, hexbin, webshot2)

conflict_prefer("select", "dplyr")
conflict_prefer("filter", "dplyr")
Show code
# Define consistent plot themes
plotTheme <- theme(
  plot.title = element_text(size = 14, face = "bold"),
  plot.subtitle = element_text(size = 10),
  plot.caption = element_text(size = 8),
  axis.text.x = element_text(size = 10, angle = 45, hjust = 1),
  axis.text.y = element_text(size = 10),
  axis.title = element_text(size = 11, face = "bold"),
  panel.background = element_blank(),
  panel.grid.major = element_line(colour = "#D0D0D0", linewidth = 0.2),
  panel.grid.minor = element_blank(),
  axis.ticks = element_blank(),
  legend.position = "right"
)

mapTheme <- theme(
  plot.title = element_text(size = 14, face = "bold"),
  plot.subtitle = element_text(size = 10),
  plot.caption = element_text(size = 8),
  axis.line = element_blank(),
  axis.text = element_blank(),
  axis.ticks = element_blank(),
  axis.title = element_blank(),
  panel.background = element_blank(),
  panel.border = element_blank(),
  panel.grid.major = element_line(colour = 'transparent'),
  panel.grid.minor = element_blank(),
  legend.position = "right",
  plot.margin = margin(1, 1, 1, 1, 'cm'),
  legend.key.height = unit(1, "cm"),
  legend.key.width = unit(0.2, "cm")
)

2.1 SEPTA Ridership Data

We use SEPTA’s Fall 2024 bus ridership summary, which represents a recent but relatively stable service period that predates the most recent service reductions. This allows us to model ridership patterns under a consistent service regime.

We first rename and retain only the variables needed for analysis and then construct a set of derived ridership metrics:

  • total_boardings – weekday + Saturday + Sunday boardings
  • total_alightings – weekday + Saturday + Sunday alightings
  • total_ridership – combined boardings and alightings
  • avg_daily_boardings / avg_daily_alightings – average daily demand across a full week
  • avg_weekday_boardings – typical weekday demand (our primary ridership measure)
  • weekend_share – share of boardings occurring on Saturday/Sunday
  • ridership_category – categorical bins describing stop activity (from “No Activity” to “Very High”)
  • log_wk_boardings / log_total_boardings – log-transformed versions for modeling

Source: SEPTA Open Data Portal

Show code
bus_ridership_raw <- st_read("./data/septa_fall_2024_bus_summary.geojson", quiet = TRUE)
Show code
# preview
glimpse(bus_ridership_raw)
Rows: 18,290
Columns: 16
$ FID       <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 1…
$ Mode      <chr> "Bus", "Bus", "Bus", "Bus", "Bus", "Bus", "Bus", "Bus", "Bus…
$ Route     <chr> "18", "18", "18", "18", "18", "18", "18", "18", "18", "18", …
$ Direction <chr> "Eastbound", "Eastbound", "Eastbound", "Eastbound", "Eastbou…
$ Route_1   <chr> "370", "371", "372", "373", "375", "467", "15421", "15424", …
$ Stop      <chr> "Upsal St & Mansfield Av", "Chew Av & Chelten Av", "Olney Av…
$ Lat       <dbl> 40.06365, 40.04550, 40.03903, 40.03887, 40.05443, 40.07590, …
$ Lon       <dbl> -75.16982, -75.16500, -75.14496, -75.14416, -75.09078, -75.0…
$ Sign_Up   <chr> "Fall 2024", "Fall 2024", "Fall 2024", "Fall 2024", "Fall 20…
$ Wk_Ons    <int> 18, 166, 668, 692, 9, 12, 228, 19, 36, 56, 48, 19, 24, 53, 8…
$ Wk_Offs   <int> 19, 175, 1163, 144, 44, 47, 0, 0, 4, 31, 11, 3, 5, 19, 11, 8…
$ Sat_Ons   <int> 13, 146, 419, 515, 4, 3, 181, 10, 19, 30, 18, 7, 9, 25, 22, …
$ Sat_Offs  <int> 9, 87, 682, 85, 24, 31, 0, 2, 5, 15, 9, 6, 8, 9, 1, 0, 15, 9…
$ Sun_Ons   <int> 12, 92, 266, 347, 4, 4, 126, 6, 16, 16, 16, 9, 10, 17, 12, 4…
$ Sun_Offs  <int> 4, 64, 455, 72, 20, 21, 0, 2, 5, 12, 4, 3, 2, 6, 5, 0, 12, 5…
$ geometry  <POINT [°]> POINT (-75.16982 40.06365), POINT (-75.165 40.0455), P…
Show code
# clean data
bus_ridership <- bus_ridership_raw %>%
  select(- Sign_Up, - Mode) %>%
  rename(fid = FID,
         route = Route,
         direction = Direction,
         stop_id = Route_1,
         stop_name = Stop,
         latitude = Lat,
         longitude = Lon,
         wk_ons = Wk_Ons,
         wk_offs = Wk_Offs,
         sat_ons = Sat_Ons,
         sat_offs = Sat_Offs,
         sun_ons = Sun_Ons,
         sun_offs = Sun_Offs)

#aggregate ridership metrics and compute average daily boardings

bus_ridership <- bus_ridership %>%
  mutate(total_boardings = wk_ons + sat_ons + sun_ons,
    total_alightings = wk_offs + sat_offs + sun_offs,
    total_ridership = total_boardings + total_alightings,
    avg_daily_boardings = (wk_ons * 5 + sat_ons + sun_ons) / 7,
    avg_daily_alightings = (wk_offs * 5 + sat_offs + sun_offs) / 7,
    avg_weekday_boardings = wk_ons / 5,
    weekend_share = (sat_ons + sun_ons) / (total_boardings + 0.001),
    # categories based on average weekday boardings
    ridership_category = case_when(
      avg_weekday_boardings == 0 ~ "No Activity",
      avg_weekday_boardings < 5 ~ "Very Low (< 5/day)",
      avg_weekday_boardings < 20 ~ "Low (5-20/day)",
      avg_weekday_boardings < 50 ~ "Medium (20-50/day)",
      avg_weekday_boardings < 100 ~ "High (50-100/day)",
      TRUE ~ "Very High (100+/day)"),
    log_wk_boardings = log(wk_ons + 1),
    log_total_boardings = log(total_boardings + 1))

glimpse(bus_ridership)
Rows: 18,290
Columns: 24
$ fid                   <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 1…
$ route                 <chr> "18", "18", "18", "18", "18", "18", "18", "18", …
$ direction             <chr> "Eastbound", "Eastbound", "Eastbound", "Eastboun…
$ stop_id               <chr> "370", "371", "372", "373", "375", "467", "15421…
$ stop_name             <chr> "Upsal St & Mansfield Av", "Chew Av & Chelten Av…
$ latitude              <dbl> 40.06365, 40.04550, 40.03903, 40.03887, 40.05443…
$ longitude             <dbl> -75.16982, -75.16500, -75.14496, -75.14416, -75.…
$ wk_ons                <int> 18, 166, 668, 692, 9, 12, 228, 19, 36, 56, 48, 1…
$ wk_offs               <int> 19, 175, 1163, 144, 44, 47, 0, 0, 4, 31, 11, 3, …
$ sat_ons               <int> 13, 146, 419, 515, 4, 3, 181, 10, 19, 30, 18, 7,…
$ sat_offs              <int> 9, 87, 682, 85, 24, 31, 0, 2, 5, 15, 9, 6, 8, 9,…
$ sun_ons               <int> 12, 92, 266, 347, 4, 4, 126, 6, 16, 16, 16, 9, 1…
$ sun_offs              <int> 4, 64, 455, 72, 20, 21, 0, 2, 5, 12, 4, 3, 2, 6,…
$ geometry              <POINT [°]> POINT (-75.16982 40.06365), POINT (-75.165…
$ total_boardings       <int> 43, 404, 1353, 1554, 17, 19, 535, 35, 71, 102, 8…
$ total_alightings      <int> 32, 326, 2300, 301, 88, 99, 0, 4, 14, 58, 24, 12…
$ total_ridership       <int> 75, 730, 3653, 1855, 105, 118, 535, 39, 85, 160,…
$ avg_daily_boardings   <dbl> 16.428571, 152.571429, 575.000000, 617.428571, 7…
$ avg_daily_alightings  <dbl> 15.4285714, 146.5714286, 993.1428571, 125.285714…
$ avg_weekday_boardings <dbl> 3.6, 33.2, 133.6, 138.4, 1.8, 2.4, 45.6, 3.8, 7.…
$ weekend_share         <dbl> 0.5813818, 0.5891075, 0.5062820, 0.5546972, 0.47…
$ ridership_category    <chr> "Very Low (< 5/day)", "Medium (20-50/day)", "Ver…
$ log_wk_boardings      <dbl> 2.944439, 5.117994, 6.505784, 6.541030, 2.302585…
$ log_total_boardings   <dbl> 3.784190, 6.003887, 7.210818, 7.349231, 2.890372…

We verified that there were no negative or missing ridership values and retained all rows. We then clipped the dataset to stops within the City of Philadelphia using Census place boundaries and projected the data to EPSG:2272.

Show code
# download Philadelphia boundary
philly_boundary <- places(state = "PA", year = 2022) %>% 
  filter(NAME == "Philadelphia") %>% 
  st_transform(4326)

# retain only boundary
philly_boundary <- philly_boundary %>% 
  st_geometry() %>% 
  st_as_sf()

# clip stops to Philadelphia boundary
ridership_philly <- st_intersection(bus_ridership, philly_boundary)

# project to EPSG:2272 (PA State Plane South)
ridership_philly <- ridership_philly %>%
  st_transform(2272)

glimpse(ridership_philly)
Rows: 10,898
Columns: 24
$ fid                   <int> 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, …
$ route                 <chr> "18", "18", "18", "18", "18", "18", "18", "18", …
$ direction             <chr> "Eastbound", "Eastbound", "Eastbound", "Eastboun…
$ stop_id               <chr> "370", "371", "372", "373", "375", "467", "15424…
$ stop_name             <chr> "Upsal St & Mansfield Av", "Chew Av & Chelten Av…
$ latitude              <dbl> 40.06365, 40.04550, 40.03903, 40.03887, 40.05443…
$ longitude             <dbl> -75.16982, -75.16500, -75.14496, -75.14416, -75.…
$ wk_ons                <int> 18, 166, 668, 692, 9, 12, 19, 36, 56, 48, 19, 24…
$ wk_offs               <int> 19, 175, 1163, 144, 44, 47, 0, 4, 31, 11, 3, 5, …
$ sat_ons               <int> 13, 146, 419, 515, 4, 3, 10, 19, 30, 18, 7, 9, 2…
$ sat_offs              <int> 9, 87, 682, 85, 24, 31, 2, 5, 15, 9, 6, 8, 9, 1,…
$ sun_ons               <int> 12, 92, 266, 347, 4, 4, 6, 16, 16, 16, 9, 10, 17…
$ sun_offs              <int> 4, 64, 455, 72, 20, 21, 2, 5, 12, 4, 3, 2, 6, 5,…
$ total_boardings       <int> 43, 404, 1353, 1554, 17, 19, 35, 71, 102, 82, 35…
$ total_alightings      <int> 32, 326, 2300, 301, 88, 99, 4, 14, 58, 24, 12, 1…
$ total_ridership       <int> 75, 730, 3653, 1855, 105, 118, 39, 85, 160, 106,…
$ avg_daily_boardings   <dbl> 16.428571, 152.571429, 575.000000, 617.428571, 7…
$ avg_daily_alightings  <dbl> 15.4285714, 146.5714286, 993.1428571, 125.285714…
$ avg_weekday_boardings <dbl> 3.6, 33.2, 133.6, 138.4, 1.8, 2.4, 3.8, 7.2, 11.…
$ weekend_share         <dbl> 0.5813818, 0.5891075, 0.5062820, 0.5546972, 0.47…
$ ridership_category    <chr> "Very Low (< 5/day)", "Medium (20-50/day)", "Ver…
$ log_wk_boardings      <dbl> 2.944439, 5.117994, 6.505784, 6.541030, 2.302585…
$ log_total_boardings   <dbl> 3.784190, 6.003887, 7.210818, 7.349231, 2.890372…
$ geometry              <POINT [US_survey_foot]> POINT (2690583 276595.2), POI…
Show code
# summary statistics
summary_stats <- ridership_philly %>%
  st_drop_geometry() %>%
  summarise(
    n_stops = n(),
    n_routes = n_distinct(route),
    total_weekday_boardings = sum(wk_ons, na.rm = TRUE),
    mean_weekday_boardings = mean(wk_ons, na.rm = TRUE),
    median_weekday_boardings = median(wk_ons, na.rm = TRUE),
    max_weekday_boardings = max(wk_ons, na.rm = TRUE),
    pct_zero_ridership = sum(wk_ons == 0) / n() * 100
  )

cat("\nSummary Statistics:\n")

Summary Statistics:
Show code
table(summary_stats) %>%
  as.data.frame() %>%
  print()
  n_stops n_routes total_weekday_boardings mean_weekday_boardings
1   10898       94                  370845        34.028720866214
  median_weekday_boardings max_weekday_boardings pct_zero_ridership Freq
1                       13                  2566   11.6902183886952    1
Show code
# Ridership by category
cat("\nStops by ridership category:\n")

Stops by ridership category:
Show code
table(ridership_philly$ridership_category) %>%
  as.data.frame() %>%
  rename(Category = Var1, Count = Freq) %>%
  mutate(Percent = round(Count / sum(Count) * 100, 1)) %>%
  print()
              Category Count Percent
1    High (50-100/day)   105     1.0
2       Low (5-20/day)  2913    26.7
3   Medium (20-50/day)   650     6.0
4          No Activity  1274    11.7
5 Very High (100+/day)    51     0.5
6   Very Low (< 5/day)  5905    54.2

The summary shows the distribution of weekday boardings and the share of stops in each ridership category.

Show code
# Map of stops by ridership category 
p1 <- ggplot(ridership_philly) +
  geom_sf(aes(color = ridership_category, size = wk_ons), alpha = 0.6) +
  scale_color_viridis_d(option = "plasma", name = "Ridership") +
  scale_size_continuous(range = c(0.5, 5), guide = "none") +
  labs(title = "Bus Stop Ridership Across Philadelphia") +
  theme_void()

p1