Case study: How do cyclists use Divvy differently?
Laura Fontanills
30 January 2023
Introduction
I completed this case study as the capstone project for my Google Data Analytics Certification.
The case study is about Chicago’s Divvy bike-share, and my objective is to evaluate how members and casual users utilize the bike-share differently, and then make recommendations to the company.
The data is taken from a real bike-share – Chicago-based Divvy – but in the original case study my objective was set for a fictional company. Because there were some differences between Divvy and the fictional Cyclistic bike-share, I decided to conduct this analysis as if I were conducting this analysis for Divvy.
Get the Raw Data
Divvy’s historical data can be accessed here. Each month’s data is contained in a single .csv file. I combined 12 months of Divvy data into a single data frame.
I conducted all cleaning, analysis, and visualization in R.
library(tidyverse) # data processing and analysis
library(lubridate) # wrangle dates and times
options(dplyr.summarise.inform = FALSE)The combined data frame has 14 variables:
Xis a row number that I’ll remove laterride_idis a unique string of letters and numbers that identifies each triprideable_typeis the type of bike used on each trip: classic, electric, or unspecifiedstarted_atcontains the date and time each trip startedended_atcontains the date and time each trip endedstart_station_namecontains the name of the dock where each trip beganstart_station_idis a unique string of letters and numbers that identifies each starting dockend_station_namecontains the name of the dock where each trip endedend_station_idis a unique string of letters and numbers that identifies each ending dockstart_latis the latitude coordinate where each trip startedstart_lngis the longitude coordinate where each trip startedend-latis the latitude coordinate where each trip endedend_lngis the longitude coordinate where each trip endedmember_casualdescribes whether a trip was taken by a Divvy member or a casual user
rides_2022 <- read.csv("rides_2022.csv")
head(rides_2022)## X ride_id rideable_type started_at ended_at
## 1 1 C2F7DD78E82EC875 electric_bike 2022-01-13 11:59:47 2022-01-13 12:02:44
## 2 2 A6CF8980A652D272 electric_bike 2022-01-10 08:41:56 2022-01-10 08:46:17
## 3 3 BD0F91DFF741C66D classic_bike 2022-01-25 04:53:40 2022-01-25 04:58:01
## 4 4 CBB80ED419105406 classic_bike 2022-01-04 00:18:04 2022-01-04 00:33:00
## 5 5 DDC963BFDDA51EEA classic_bike 2022-01-20 01:31:10 2022-01-20 01:37:12
## 6 6 A39C6F6CC0586C0B classic_bike 2022-01-11 18:48:09 2022-01-11 18:51:31
## start_station_name start_station_id end_station_name
## 1 Glenwood Ave & Touhy Ave 525 Clark St & Touhy Ave
## 2 Glenwood Ave & Touhy Ave 525 Clark St & Touhy Ave
## 3 Sheffield Ave & Fullerton Ave TA1306000016 Greenview Ave & Fullerton Ave
## 4 Clark St & Bryn Mawr Ave KA1504000151 Paulina St & Montrose Ave
## 5 Michigan Ave & Jackson Blvd TA1309000002 State St & Randolph St
## 6 Wood St & Chicago Ave 637 Honore St & Division St
## end_station_id start_lat start_lng end_lat end_lng member_casual
## 1 RP-007 42.01280 -87.66591 42.01256 -87.67437 casual
## 2 RP-007 42.01276 -87.66597 42.01256 -87.67437 casual
## 3 TA1307000001 41.92560 -87.65371 41.92533 -87.66580 member
## 4 TA1309000021 41.98359 -87.66915 41.96151 -87.67139 casual
## 5 TA1305000029 41.87785 -87.62408 41.88462 -87.62783 member
## 6 TA1305000034 41.89563 -87.67207 41.90312 -87.67394 memberData Cleaning
Considerations
Before working in R, I reviewed each month’s data independently in Excel. I discovered the following inconsistencies that I would need to address.
Test rides outside the Chicago area or with no geolocation
Rides starting before and ending after the DST time changes in March and November
Many rides shorter than 1 minute, which could be due to bike problems
Many rides longer than 24 hours, which could be due to docking problems
Rides with no end location
A nonspecific “docked” bike type associated with a small percentage of rides by casual users
I confirmed that all rows are unique by checking the number of ride IDs. TRUE indicates that all rows are unique.
n_distinct(rides_2022$ride_id) == nrow(rides_2022)## [1] TRUEI also confirmed that all rides were taken by either members or casual users.
rides_2022 %>%
group_by(member_casual) %>%
summarize(num_rides = n()) %>%
mutate(pct_rides = 100*(num_rides/sum(num_rides))) # calculate percentage## # A tibble: 2 × 3
## member_casual num_rides pct_rides
## <chr> <int> <dbl>
## 1 casual 2322032 41.0
## 2 member 3345685 59.0Ride Time Data
I transformed the started_at and ended_at
fields from character vectors to datatime.
rides_2022$started_at <- ymd_hms(rides_2022$started_at)
rides_2022$ended_at <- ymd_hms(rides_2022$ended_at)I create a new column called ride_length for ride
duration in minutes. Some rides have negative ride duration, so I’ll
need to filter these out.
rides_2022$ride_length <- as.numeric(rides_2022$ended_at - rides_2022$started_at)
rides_2022$ride_length <- rides_2022$ride_length / 60
rides_2022 %>%
filter(ride_length < 0) %>%
select(started_at, ended_at, ride_length) %>%
head()## started_at ended_at ride_length
## 1 2022-03-05 11:00:57 2022-03-05 10:55:01 -5.9333333
## 2 2022-03-05 11:38:04 2022-03-05 11:37:57 -0.1166667
## 3 2022-05-30 11:06:29 2022-05-30 11:06:17 -0.2000000
## 4 2022-06-07 19:15:39 2022-06-07 17:05:37 -130.0333333
## 5 2022-06-07 19:14:46 2022-06-07 17:07:45 -127.0166667
## 6 2022-06-23 19:22:57 2022-06-23 19:21:46 -1.1833333121089 rides are very short (less than 1 minute), and of these 431 are 0 seconds long.
5360 rides are very long (greater than 24 hours). With a day pass, Divvy lets you ride up to 3 hours on a single trip before incurring a fee, and the pass is good for 24 hours. It’s impossible to know whether trips longer than 3 hours were real trips or whether the user had docking issues. 19592 rides are longer than 3 hours. For my analysis I considered typical use to be between 60 seconds and 3 hours.
Ride Location and Bike Data
There are three bike types in the data frame - classic_bike, electric_bike, and docked_bike. Divvy docks use a different geolocation system than the bikes themselves– docks have geolocation with 0.00001° sensitivity, while undocked bikes have geolocation with 0.01° sensitivity. Many Divvy docks are within 0.01° of other docks, especially in the city center where most rides take place, so undocked bikes can’t reliably be matched with nearby docks. Divvy allows their users to take out and drop off undocked bikes for a fee.
# How many rides were taken by members, casual users on each type
rides_2022 %>%
group_by(member_casual, rideable_type) %>%
summarize(num_rides = n()) %>%
mutate(pct_rides = 100*(num_rides/sum(num_rides))) # calculate percentage## # A tibble: 5 × 4
## # Groups: member_casual [2]
## member_casual rideable_type num_rides pct_rides
## <chr> <chr> <int> <dbl>
## 1 casual classic_bike 891459 38.4
## 2 casual docked_bike 177474 7.64
## 3 casual electric_bike 1253099 54.0
## 4 member classic_bike 1709755 51.1
## 5 member electric_bike 1635930 48.9177474 rides were taken on the nonspecific docked_bike type. Only casual riders used the docked_bike type, which accounted for 7.6% of all casual rides. These bikes were undocked from 678 unique Divvy stations during all months of the year.
# When were docked bikes taken out, returned
rides_2022 %>%
filter(rideable_type == "docked_bike") %>%
group_by(rideable_type) %>%
summarize(min_start = min(started_at),
max_start = max(started_at),
min_end = min(ended_at),
max_end = max(ended_at))## # A tibble: 1 × 5
## rideable_type min_start max_start min_end
## <chr> <dttm> <dttm> <dttm>
## 1 docked_bike 2022-01-01 00:24:47 2022-12-31 23:44:47 2022-01-01 00:26:19
## # … with 1 more variable: max_end <dttm># Where were docked bikes taken out
rides_2022 %>%
filter(rideable_type == "docked_bike") %>%
group_by(start_station_name) %>%
summarize(num_rides=n()) %>%
# arrange(num_rides)
arrange(desc(num_rides))## # A tibble: 688 × 2
## start_station_name num_rides
## <chr> <int>
## 1 Streeter Dr & Grand Ave 11788
## 2 DuSable Lake Shore Dr & Monroe St 6969
## 3 Millennium Park 5796
## 4 Shedd Aquarium 4658
## 5 Michigan Ave & Oak St 4196
## 6 Dusable Harbor 3163
## 7 DuSable Lake Shore Dr & North Blvd 3111
## 8 Adler Planetarium 2698
## 9 Theater on the Lake 2627
## 10 Montrose Harbor 2363
## # … with 678 more rowsAll rides had starting geolocations in the start_lat and start_lng fields. 833064 rides had no starting station information. All of these bikes were electric_bike type. ` Not all rides had starting geolocations in the end_lat and end_lng fields. 892742 rides had no ending station information. These bikes were classic_bike and electric_bike types.
I located 1 ride outside the city limits, which I identified as a test ride, and 8 rides to Green St & Madison Ave, which is not listed as a station (current or former) on Divvy’s website. I will remove these from the data set during cleaning.
# set city limits
city_lat <- c(41, 43)
city_lng <- c(-89, -87)
# find rides outside city area
rides_2022 %>%
filter((!between(start_lat, city_lat[1], city_lat[2])) |
(!between(start_lng, city_lng[1], city_lng[2]))) %>%
select(ride_id, start_station_name, start_lat, start_lng)## ride_id start_station_name start_lat
## 1 3327172413547F64 Pawel Bialowas - Test- PBSC charging station 45.63503
## start_lng
## 1 -73.79648rides_2022 %>%
filter((!between(end_lat, city_lat[1], city_lat[2])) |
(!between(end_lng, city_lng[1], city_lng[2])) & end_lat != 0 & end_lng != 0) %>%
select(ride_id, end_station_name, end_lat, end_lng)## ride_id end_station_name end_lat end_lng
## 1 3B47B333C0D186F0 Green St & Madison Ave* 0 0
## 2 42AF82C53D831251 Green St & Madison Ave* 0 0
## 3 BB8AA29838266294 Green St & Madison Ave* 0 0
## 4 6AFE1471227BD76F Green St & Madison Ave* 0 0
## 5 E9495F1DC3475D41 Green St & Madison Ave* 0 0
## 6 75DE33501313D0CE Green St & Madison Ave* 0 0
## 7 0A6988FE859F4D54 Green St & Madison Ave* 0 0
## 8 7F49424E860E7094 Green St & Madison Ave* 0 0I saved the cleaned and transformed data to a new data frame.
# create new data frame - clean
rides_clean <-rides_2022 %>%
# keep rides b/w 1 min, 3 hr
filter(ride_length > 1) %>%
filter(ride_length < 60*3) %>%
# filter out rides w/o location data
filter(!is.na(start_lat)) %>%
filter(!is.na(end_lat)) %>%
# filter out rides during DST change on 11/6
filter(!(started_at < '2022-11-06 02:00:00' & ended_at > '2022-11-06 02:00:00')) %>%
# filter out rides not in Chicago
filter(between(start_lat, city_lat[1], city_lat[2])) %>%
filter(between(end_lat, city_lat[1], city_lat[2])) %>%
# select columns for new data frame
select(ride_id, member_casual, rideable_type, started_at, ended_at, start_station_name, end_station_name, start_lat, start_lng, end_lat, end_lng, ride_length)
head(rides_clean)## ride_id member_casual rideable_type started_at
## 1 C2F7DD78E82EC875 casual electric_bike 2022-01-13 11:59:47
## 2 A6CF8980A652D272 casual electric_bike 2022-01-10 08:41:56
## 3 BD0F91DFF741C66D member classic_bike 2022-01-25 04:53:40
## 4 CBB80ED419105406 casual classic_bike 2022-01-04 00:18:04
## 5 DDC963BFDDA51EEA member classic_bike 2022-01-20 01:31:10
## 6 A39C6F6CC0586C0B member classic_bike 2022-01-11 18:48:09
## ended_at start_station_name
## 1 2022-01-13 12:02:44 Glenwood Ave & Touhy Ave
## 2 2022-01-10 08:46:17 Glenwood Ave & Touhy Ave
## 3 2022-01-25 04:58:01 Sheffield Ave & Fullerton Ave
## 4 2022-01-04 00:33:00 Clark St & Bryn Mawr Ave
## 5 2022-01-20 01:37:12 Michigan Ave & Jackson Blvd
## 6 2022-01-11 18:51:31 Wood St & Chicago Ave
## end_station_name start_lat start_lng end_lat end_lng
## 1 Clark St & Touhy Ave 42.01280 -87.66591 42.01256 -87.67437
## 2 Clark St & Touhy Ave 42.01276 -87.66597 42.01256 -87.67437
## 3 Greenview Ave & Fullerton Ave 41.92560 -87.65371 41.92533 -87.66580
## 4 Paulina St & Montrose Ave 41.98359 -87.66915 41.96151 -87.67139
## 5 State St & Randolph St 41.87785 -87.62408 41.88462 -87.62783
## 6 Honore St & Division St 41.89563 -87.67207 41.90312 -87.67394
## ride_length
## 1 2.950000
## 2 4.350000
## 3 4.350000
## 4 14.933333
## 5 6.033333
## 6 3.366667This file has 5525383 lines and 12 columns. It is ready to be analyzed.
Analysis and Visualizations
Finding 1: Casual users take longer trips than members
Casual trips are about 8 minutes longer, on average, than member trips. Perhaps they’re not as used to bikes or bike-shares, or perhaps they are using the bikes for fun.
tapply(rides_clean$ride_length, rides_clean$member_casual, summary)## $casual
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.017 7.600 13.200 20.523 24.117 179.983
##
## $member
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.017 5.317 9.017 12.265 15.450 179.900length_plot <-
rides_clean %>%
group_by(member_casual) %>%
summarize(number_of_rides = n(), mean_length = mean(ride_length)) %>%
arrange(member_casual) %>%
ggplot(aes(x=member_casual, y=mean_length, fill=member_casual)) +
geom_col(position = "dodge") +
labs(
title = "Average ride duration",
subtitle = "For all trips between 1 minute and 3 hours long",
x = "User type",
y = "Average ride duration (minutes)",
fill = "User type") +
scale_y_continuous(labels = scales::comma) +
theme_classic() +
scale_fill_brewer(palette = "Blues", direction = -1)
length_plot
Finding 2: Casual riders take more weekend trips
Member rides peak during the work week, which could indicate that many members use bikes for commuting. Casual rides peak on Saturday and Sunday, suggesting that casual users use bikes for leisure or tourism.
weekday_plot <-
rides_clean %>%
mutate(weekday = wday(started_at, label = TRUE, abbr = FALSE)) %>%
group_by(member_casual, weekday) %>%
summarize(count=n()) %>%
ggplot(aes(x=weekday, y=count, fill=member_casual))+
geom_col(position="dodge") +
labs(
title = "Number of rides by weekday",
subtitle = "For all trips between 1 minute and 3 hours long",
x = "User type",
y = "Number of rides",
fill = "User type") +
scale_y_continuous(labels = scales::comma) +
theme_classic() +
scale_fill_brewer(palette = "Blues", direction = -1)
weekday_plot
## Finding 3: Casual riders use a few stations relatively frequently
Divvy has 678 stations at the time of writing. Casual riders show a clear preference for 5 stations – these stations are where 8% of all casual rides begin. The top 5 casual starting stations are located in parks in the city center, close to the waterfront. These are very popular locations for tourists and locals.
Members don’t especially favor any stations, and no station accounts for more than 1% of all member rides.
rides_clean %>%
filter(member_casual == "casual") %>%
group_by(member_casual, start_station_name) %>%
summarize(num_rides = n()) %>%
arrange(desc(num_rides)) %>%
mutate(percent_rides = 100*(num_rides/sum(num_rides))) %>%
head(10)## # A tibble: 10 × 4
## # Groups: member_casual [1]
## member_casual start_station_name num_rides percent_rides
## <chr> <chr> <int> <dbl>
## 1 casual "" 331467 14.7
## 2 casual "Streeter Dr & Grand Ave" 56441 2.50
## 3 casual "DuSable Lake Shore Dr & Monroe St" 30971 1.37
## 4 casual "Millennium Park" 24709 1.09
## 5 casual "Michigan Ave & Oak St" 24618 1.09
## 6 casual "DuSable Lake Shore Dr & North Blvd" 22962 1.02
## 7 casual "Shedd Aquarium" 19708 0.873
## 8 casual "Theater on the Lake" 18054 0.800
## 9 casual "Wells St & Concord Ln" 15913 0.705
## 10 casual "Dusable Harbor" 13689 0.606 rides_clean %>%
filter(member_casual == "member") %>%
group_by(member_casual, start_station_name) %>%
summarize(num_rides = n()) %>%
arrange(desc(num_rides)) %>%
mutate(percent_rides = 100*(num_rides/sum(num_rides))) %>%
head(10)## # A tibble: 10 × 4
## # Groups: member_casual [1]
## member_casual start_station_name num_rides percent_rides
## <chr> <chr> <int> <dbl>
## 1 member "" 469193 14.4
## 2 member "Kingsbury St & Kinzie St" 24483 0.749
## 3 member "Clark St & Elm St" 21579 0.660
## 4 member "Wells St & Concord Ln" 20937 0.641
## 5 member "University Ave & 57th St" 19487 0.596
## 6 member "Clinton St & Washington Blvd" 19351 0.592
## 7 member "Ellis Ave & 60th St" 19088 0.584
## 8 member "Wells St & Elm St" 18650 0.571
## 9 member "Loomis St & Lexington St" 18638 0.570
## 10 member "Clinton St & Madison St" 18466 0.565Finding 4: Casual riders use more e-bikes
In 2022, Casual riders used more electric bikes than classic bikes. By contrast, members favored classic bikes.
rides_clean %>%
group_by(member_casual, rideable_type) %>%
summarize(num_rides = n()) %>%
mutate(pct_rides = 100* num_rides/sum(num_rides))## # A tibble: 5 × 4
## # Groups: member_casual [2]
## member_casual rideable_type num_rides pct_rides
## <chr> <chr> <int> <dbl>
## 1 casual classic_bike 870917 38.6
## 2 casual docked_bike 168568 7.47
## 3 casual electric_bike 1218180 54.0
## 4 member classic_bike 1681261 51.5
## 5 member electric_bike 1586457 48.5bike_type_plot <- rides_clean %>%
group_by(rideable_type, member_casual) %>%
filter(rideable_type == "classic_bike" | rideable_type == "electric_bike") %>%
summarize(num_rides = n()) %>%
arrange(member_casual) %>%
ggplot(aes(x=member_casual, y=num_rides, fill=rideable_type)) +
geom_col(position = "dodge") +
labs(
title = "Number of rides with Classic Bikes, Electric Bikes",
subtitle = "For all trips between 1 minute and 3 hours long",
x = "Membership type",
y = "Number of rides",
fill = "Bike type",
labels = c("classic_bike", "electric_bike")) +
scale_y_continuous(labels = scales::comma) +
theme_classic() +
scale_fill_brewer(palette = "Blues")
bike_type_plot
Recommendations
Appeal to electric bike users.
Casual riders take more trips on e-bikes than on classic bikes. Electric bikes are fun, easy to use, and somewhat novel. Communicating the advantages of membership for e-bike users could convert some casual users.
Target casual cyclists riding for leisure.
Casual riders take more trips on weekends. Advertising membership as a cost-effective way of exploring the city could resonate with people who don’t have time to cycle during the week, and who want to use bikes for fun. Weekend or off-peak memberships could be an option to explore.
Increase advertising at high-traffic stations.
There are 5 stations that are heavily used by casual riders, all of which are located in the city center. Advertise membership to locals who want to get around Chicago to see parks, museums, and restaurants.