How much of Denver can walk to frequent transit?

A SQL and R analysis of RTD’s weekday service, from the morning peak to late night

Author

Austin Foster

Published

October 8, 2026

Executive summary

On a typical weekday (Wednesday, September 30, 2026), RTD ran 128 routes. Counting a route as frequent when its median gap between trips is 15 minutes or less in both directions, 29 routes were frequent in the AM peak (6–9 AM), 25 at mid-day, and 28 in the PM peak, but only 5 between 7 PM and 3 AM. Of the 7,453 stops served that day, 22.4% had frequent service in the AM peak and 4.8% late at night.

About 13.2% of residents in the seven RTD counties, and 33.1% of households without a car, live within a quarter-mile walk of a frequent stop in the AM peak. After 7 PM that falls to 4.1% and 15.5%.

Why this matters

Frequency is what turns a bus line into something people can use without planning around it. When a bus comes every 15 minutes or better, riders can walk to the stop and go, a missed trip costs minutes rather than an hour, and transfers between routes become practical. Service that runs less often forces people to build their day around a timetable, which is a burden most drivers never face. That burden falls hardest on people without a car, who rely on transit for work, school, groceries, and appointments, and whose trips don’t all happen at rush hour. Measuring where frequent service exists, and when it disappears, shows which neighborhoods have transit they can count on all day and which have it only at the peaks.

Data

Source What it provides Version
RTD GTFS feed Every scheduled trip, stop, and departure time Downloaded 2026-10-06
ACS 5-year estimates (tables B01003, B25044) Population and households without a vehicle, by block group 2019–2023
OpenStreetMap via Geofabrik Streets, sidewalks, and paths for walking distances Downloaded with the GTFS run

Method

  1. Load the GTFS text files into DuckDB, a SQL database that runs inside R.
  2. Pick a typical weekday. GTFS covers months of service, so the SQL expands the calendar into individual dates, applies holiday exceptions, and chooses the Wednesday with the most scheduled trips.
  3. Split the day into four periods: AM Peak (6–9 AM), Mid-Day (9 AM–3 PM), PM Peak (3–7 PM), and Late Night (7 PM–3 AM). GTFS writes after-midnight times as 24:00 or later, so Late Night ends at 27:00. Trips between 3 and 6 AM fall outside the four periods.
  4. Measure route headways (sql/03_route_headways.sql). For each route and direction, a LAG() window function finds the time between successive trips at the stop the route serves most often. Only gaps between two trips that both leave inside the same period count. A gap that runs across a period boundary, or from the last trip of the night to 3 AM, is left out. A route is frequent in a period if its median headway is 15 minutes or less in every direction, with at least 4 trips each way (an hour’s worth at 15 minutes). The median is the typical scheduled gap: half the gaps are shorter, half are longer, so one odd gap at a service change doesn’t decide the result.
  5. Flag frequent stops (sql/04_stop_frequency.sql). The same calculation runs at every stop. A stop is frequent in a period if a single route-direction serving it has a median headway of 15 minutes or less there, with at least 4 departures. Several infrequent routes stacked together don’t count, because riders can’t combine them into one wait.
  6. Build walksheds on the street network (R/04_osm_network.R, R/05_walksheds.R). OpenStreetMap streets, sidewalks, and paths become a routable network. One shortest-path search per period finds everywhere within a quarter-mile walk of a frequent stop, so rail lines, highways, and dead ends limit access the way they do on foot.
  7. Estimate who lives nearby (R/06_census_access.R). ACS block groups are counted in proportion to how much of their area falls inside each period’s walkshed.

Results

Service by time of day

For each route running in a period, the SQL takes the median gap between its trips in its less frequent direction. Table 1 reports the median of those values across the whole system, so every route counts once however often it runs. Half of RTD’s routes come at least this often.

Code
m |>
  transmute(
    Period = PERIOD_LABELS[period],
    `Median headway (min)` = fmt_min(system_median_headway_min)
  ) |>
  knitr::kable(align = "lr")
Table 1: Median scheduled headway across all RTD routes, by time of day
Period Median headway (min)
AM Peak (6-9 AM) 30
Mid-Day (9 AM-3 PM) 30
PM Peak (3-7 PM) 30
Late Night (7 PM-3 AM) 56.3

How often do routes run?

Code
hw <- routes |>
  filter(!is.na(median_headway_min)) |>
  mutate(
    median_headway_min = pmin(median_headway_min, 120),
    period = factor(PERIOD_LABELS[period], levels = PERIOD_LABELS)
  )

# The threshold label ends at y = 51.5 in every panel.
threshold_lab <- tibble(period = factor(PERIOD_LABELS, levels = PERIOD_LABELS),
                        y = 51.5)

ggplot(hw, aes(median_headway_min, fill = mode)) +
  geom_histogram(binwidth = 5, boundary = 0, color = "white", linewidth = 0.2) +
  geom_vline(xintercept = 15, linetype = "dashed") +
  geom_text(data = threshold_lab, aes(x = 15, y = y), inherit.aes = FALSE,
            label = "Freq. Service Limit", angle = 90,
            hjust = 1, vjust = -0.5, size = 3.2, fontface = "italic") +
  # scales = "free_x" draws an x-axis under every panel; the fixed limits
  # keep all four panels on the same 0-120+ scale so they stay comparable
  facet_wrap(~ period, ncol = 2, scales = "free_x") +
  scale_x_continuous(limits = c(0, 125), breaks = seq(0, 120, 15),
                     labels = c(seq(0, 105, 15), "120+"), expand = expansion(mult = 0.01)) +
  # headroom above the tallest bars so the vertical label fits
  scale_y_continuous(limits = c(0, 55), breaks = seq(0, 50, 10), expand = expansion(mult = 0)) +
  scale_fill_viridis_d(option = "viridis", end = 0.9) +
  labs(x = "Median headway (minutes)", y = "Number of routes", fill = NULL) +
  theme_minimal(base_size = 12) +
  theme(legend.position = "top",
        strip.text = element_text(face = "bold", margin = margin(b = 10)),
        axis.title = element_text(face = "bold", size = 13),
        axis.title.x = element_text(margin = margin(t = 12)),
        axis.title.y = element_text(margin = margin(r = 12)),
        panel.spacing.x = unit(2.5, "lines"),   # room between the two columns of panels
        panel.spacing.y = unit(1.8, "lines"),
        panel.grid.minor = element_blank())
Figure 1: Median headway by route and time of day (weaker direction). Routes at or left of the dashed line run every 15 minutes or better. Routes with a single trip in a period have no headway and are not shown.

The frequent routes

Each cell shows the route’s median headway in that period, in its less frequent direction, using only gaps between trips that leave inside the period. Bold marks periods where the route is frequent; a dash means no service (or a single trip) in that period. The shortest and longest gaps for each route and period are in Additional Data.

Code
fmt <- function(x) ifelse(x == round(x), sprintf("%.0f", x), sprintf("%.1f", x))
frequent_ids <- routes |> filter(is_frequent) |> distinct(route_id)
routes |>
  semi_join(frequent_ids, by = "route_id") |>
  filter(!is.na(median_headway_min)) |>
  mutate(cell = fmt(median_headway_min),
         # asterisk: the median isn't a standard clock-face headway
         cell = ifelse(median_headway_min %in% STANDARD_HEADWAYS, cell, paste0(cell, "\\*")),
         cell = ifelse(is_frequent, paste0("**", cell, "**"), cell),
         period = period_factor(period)) |>
  group_by(route_id) |>
  mutate(n_frequent = sum(is_frequent), best = min(median_headway_min)) |>
  ungroup() |>
  select(route_id, Route = route_short_name, Name = route_long_name, Mode = mode,
         n_frequent, best, period, cell) |>
  pivot_wider(names_from = period, values_from = cell, values_fill = "–") |>
  arrange(desc(n_frequent), best) |>
  select(Route, Name, Mode, any_of(PERIODS)) |>
  knitr::kable(align = "lllrrrr")
Table 2: Routes that are frequent in at least one period: median headway in minutes
Route Name Mode AM Peak Mid-Day PM Peak Late Night
FreeRide 16th Street FreeRide Bus 3 3 3 6.8*
31 Federal Blvd Bus 10 10 10 15
SKIP Broadway Bus 10 10 10 15
15 East Colfax Avenue Bus 10 10 10 15
0 Broadway Bus 15 15 15 15
43 Martin Luther King Blvd Bus 7.5 15 8* 30
15L East Colfax Limited Bus 10 10 10 30
16 West Colfax Avenue Bus 15 10 10 30
0L South Broadway Limited Bus 15 15 15 –
Bound 30th Street Bus 15 15 15 30
E Union Station to RidgeGate Parkway Station Light rail 15 15 15 30
30 South Federal Blvd Bus 15 15 15 30
17 Golden/Colfax Bus 15 15 15 30
JUMP Boulder / Lafayette/ Erie Bus 15 15 15 30
105 Havana Street Bus 15 15 15 30
21 East Evans Avenue Bus 15 15 15 30
FF1 Flatiron Flyer Bus 15 15 15 30
153 Chambers Road Bus 15 15 15 30
G Union Station to Wheat Ridge Ward Station Commuter rail 15 15 15 30
W Union Station to Jefferson County Government Center-Golden Station Light rail 15 15 15 30
C Union Station to Mineral Station Light rail 15 15 15 30
T Lincoln Station to Alameda Station Light rail 15 15 15 30
40 Colorado Boulevard Bus 15 15 15 30
73 Quebec Street Bus 15 15 15 30
R RidgeGate Parkway Station to Peoria Station Light rail 15 15 15 30
MetroRide Free MetroRide Bus 10 – 10 –
121 Peoria Street Bus 15 15.6* 15 30
120X Wagon Road / Thornton Express Bus 15 30 17* 40*
12 South Downing Bus 15 30 15.5* 30
204 Table Mesa / Moorhead / North 19th Bus 16* 30 15 30

* The median isn’t a standard clock-face headway (3, 4, 5, 6, 7.5, 10, 12, 15, 20, 30, 45, 60, or 90 minutes), so riders won’t see the same minutes past the hour all period. Usually the route mixes two headways in the period, such as alternating 7 and 8 minutes or switching between 15 and 30. See Additional Data for each route’s shortest and longest gap.

Where is frequent service?

Code
library(leaflet)
library(sf)
period_cols <- setNames(viridisLite::viridis(4, end = 0.9), PERIODS)

stop_popup <- stops |>
  mutate(txt = case_when(
    routes_in_period == 0         ~ "no service",
    is.na(best_median_headway_min) ~ "fewer than 4 trips",
    TRUE                           ~ paste0(best_median_headway_min, " min"))) |>
  group_by(stop_id) |>
  summarise(popup = paste0("<b>", htmltools::htmlEscape(first(stop_name)), "</b><br>",
                           "<i>Best route, median headway</i><br>",
                           paste0(period, ": ", txt, collapse = "<br>")), .groups = "drop")
all_stops <- stops |> distinct(stop_id, stop_lat, stop_lon) |> left_join(stop_popup, by = "stop_id")

map <- leaflet() |>
  # Keyless tile layers. (CARTO basemaps started requiring an API key in 2026,
  # which is why the old map showed "API key required".)
  addProviderTiles(providers$Esri.WorldGrayCanvas, group = "Light") |>
  addProviderTiles(providers$Esri.WorldImagery,    group = "Aerial imagery") |>
  addProviderTiles(providers$OpenStreetMap,        group = "Streets") |>
  addCircleMarkers(data = all_stops, lng = ~stop_lon, lat = ~stop_lat,
                   radius = 1.5, color = "#8C8C8C", stroke = FALSE, fillOpacity = 0.5,
                   popup = ~popup, group = "All stops")

if (has_walksheds) {
  sheds <- st_read(walkshed_file, quiet = TRUE) |>
    filter(walkshed %in% PERIODS) |>
    st_transform(PROJECTED_CRS) |>
    st_simplify(dTolerance = 15, preserveTopology = TRUE) |>   # lighter web page
    st_transform(4326)
}

for (p in PERIODS) {
  grp <- PERIOD_LABELS[[p]]
  shed_p <- if (has_walksheds) filter(sheds, walkshed == p, !st_is_empty(geometry)) else NULL
  if (!is.null(shed_p) && nrow(shed_p) > 0) {
    map <- map |>
      addPolygons(data = shed_p, group = grp,
                  color = period_cols[[p]], weight = 1, fillOpacity = 0.25,
                  options = pathOptions(interactive = FALSE))   # clicks pass through to stops
  }
  map <- map |>
    addCircleMarkers(data = filter(stops, period == p, is_frequent) |>
                       left_join(stop_popup, by = "stop_id"),
                     lng = ~stop_lon, lat = ~stop_lat, radius = 3.5,
                     color = period_cols[[p]], stroke = FALSE, fillOpacity = 0.9,
                     popup = ~popup, group = grp)
}

map |>
  addLegend(colors = c(unname(period_cols), "#8C8C8C"),
            labels = c(unname(PERIOD_LABELS), "Other stops"),
            title = "Frequent service", opacity = 1) |>
  addLayersControl(baseGroups = c("Light", "Aerial imagery", "Streets"),
                   overlayGroups = c(unname(PERIOD_LABELS), "All stops"),
                   options = layersControlOptions(collapsed = TRUE)) |>
  hideGroup(unname(PERIOD_LABELS[-1]))     # start with AM Peak only
Figure 2: Frequent stops and their quarter-mile walksheds by time of day. Use the layer button (top right) to switch periods or turn on aerial imagery. Click a stop for its headways.

Who lives near frequent service?

A worker on a 9-to-5 schedule only needs the peaks. Someone who heads to work in the afternoon and comes home at midnight needs frequent service in both directions of that trip, so their access is capped by the Late Night row.

Table 3: Residents and households without a car within a quarter-mile walk of a frequent stop, by time of day
Period Residents Share of residents Households without a car Share of car-free households
AM Peak (6-9 AM) 429,126 13.2% 25,486 33.1%
Mid-Day (9 AM-3 PM) 384,219 11.8% 24,065 31.3%
PM Peak (3-7 PM) 425,469 13.1% 25,327 32.9%
Late Night (7 PM-3 AM) 132,555 4.1% 11,933 15.5%
All four periods 132,554 4.1% 11,933 15.5%

For comparison, the straight-line quarter-mile circles used in an earlier version of this analysis would have counted 16.4% of residents in the AM peak instead of 13.2%. Measuring along the street network matters most where freeways, rail lines, and cul-de-sacs sit between people and the stop.

Limitations

  • Walk network. OpenStreetMap is very complete in metro Denver, but it often draws a street with sidewalks as one centerline, so the analysis assumes every street is walkable on both sides and ignores how hard it is to cross a wide arterial.
  • Areal weighting assumes people are spread evenly across each block group.
  • One day, scheduled service. Results describe scheduled service on one Wednesday, not weekends or actual on-time performance.
  • The median describes the typical gap, not the worst one. A route that runs every 15 minutes for most of a period and every 30 for the rest counts as frequent, though riders arriving during the 30-minute stretch wait longer. The min–max range in the frequent-routes table shows when that happens.
  • Gaps are measured only inside each period. A route that runs every 10 minutes from 6 to 7 AM and then stops still counts as frequent for the AM Peak, because the wait after its last trip isn’t counted. The 4-trip minimum keeps a pair of closely spaced trips (such as a school tripper) from qualifying on its own.
  • Route headways are measured at one stop (the one each route serves most often). The stop-level results use actual departure times at every stop, so they also capture short-turn trips that never reach that stop.
  • Late Night crosses midnight. Trips that leave between midnight and 3 AM on the service day are treated as the end of the previous night’s service.

Reproduce this

git clone https://github.com/austinthomas7/rtd-frequent-network

Then run R/00_setup.R once, run_all.R, and quarto render. All code is on GitHub.