Cyclistic Bike-Share Analysis

Project background:

• This is a case study project made as part of the Google Data Analytics course.
• For a business stakeholder perspective, a PowerPoint version (with speaker notes) is available in this LINK.
  (Press CTRL + Click on the link to open in a new tab).

Setting up my environment & data:

library(tidyverse)
library(RColorBrewer)
library(scales)
library(here)

tripdata_summary_weather <- read_csv(here("processed_data/r", "tripdata_summary_weather.csv"))
tripdata_summary_daily <- read_csv(here("processed_data/r", "tripdata_summary_daily.csv"))
tripdata_time_of_day <- read_csv(here("processed_data/r", "tripdata_time_of_day.csv"))
tripdata_stations <- read_csv(here("processed_data/r", "tripdata_stations.csv"))
tripdata_summary <- read_csv(here("processed_data/r", "tripdata_summary.csv"))
station_names <- read_csv(here("processed_data/r", "station_names.csv"))

Business tasks:

• How do annual members and casual riders use Cyclistic bikes differently?
• Identify and compare bike trip trends in Chicago, Illinois to provide insights in designing a marketing strategy to convert “Casual” riders into an annual “Member” and increase revenue.

Presentation objectives:

• Share an overview of historical bike trips data.
• Examine the seasonal trends using historical data.
• Discuss any potential areas for further exploration.

The data story:

      Bike trips trend

• The data collected is from April ’20 until August ’21 (17 months).
• Member rider trips are generally higher but Casual riders increased and overtaken starting June ’21.
• Overall, trips increases from April to August and trends down from September to February.

tripdata_summary %>%
  ggplot(aes(x = s_year_month, y = ride_count, color = member_casual)) +
  theme_grey(base_size = 12) +
  geom_smooth(se = FALSE, size = 1.1) +
  scale_x_date(date_breaks = "1 month", date_labels = "%b %y") +
  scale_color_manual(values=c("purple","gold")) +
  scale_y_continuous(labels = scales::comma) +
  labs(x = "Month-Year", y = "Ride Count", color  = "Member Type") +
  theme(legend.position = "bottom")

head(tripdata_summary)

## # A tibble: 6 x 4
##   member_casual s_year_month mean_ride_length ride_count
##   <chr>         <date>                  <dbl>      <dbl>
## 1 casual        2020-04-01               72.5      23553
## 2 casual        2020-05-01               50.6      86635
## 3 casual        2020-06-01               51.3     154186
## 4 casual        2020-07-01               59.3     267977
## 5 casual        2020-08-01               44.5     280498
## 6 casual        2020-09-01               38.5     214102

      Total trip count of riders

• An overall total of 6,005,431 bike trips during the 17-month observation period.
• There is a 22% difference of trips between Casual and Member riders.
• Due to the high trip count in the Casual category, there is still a big opportunity for member conversion.

tripdata_summary %>% 
  group_by(member_casual) %>% 
  summarise(total_ride = sum(ride_count)) %>% 
  ggplot(aes(x = member_casual, y = total_ride, fill = member_casual)) +
  theme_grey(base_size = 12) +
  theme(axis.title.x=element_blank()) +
  geom_col(width = 0.50, show.legend = FALSE) +
  scale_y_continuous(labels = scales::comma) +
  scale_fill_manual(values=c("purple", "gold")) +
  stat_summary(fun = "sum", aes(label = format(..y.., big.mark = ",")),
               geom = "text", vjust = 2, size = 3.5) +
  ylab("Ride Count") +
  scale_x_discrete(labels = c('Casual', 'Member'))

      Seasonal trips trend

• In the hotter months, trips tend to rise. This can be related to increased leisure activities outdoors.
• In the colder months, trips tend to fall. Travelling through bike means being exposed in rain and snow.
  Generally, people avoid getting wet or sustain accidents on a slippery road.

tripdata_summary_weather %>% 
  ggplot(aes(x = s_year_month, y = ride_count)) +
  theme_dark(base_size = 12) +
  geom_smooth(color = "gray", se = FALSE, size = 1) +
  geom_point(aes(color = avg_temp), size = 3) +
  scale_x_date(date_breaks = "1 month", date_labels = "%b %y") +
  scale_y_continuous(labels = scales::comma) +
  scale_color_distiller(name = "Temp (C)", palette = "RdBu") +
  labs(x = "Month-Year", y = "Ride Count")

head(tripdata_summary_weather)

## # A tibble: 6 x 4
##   s_year_month avg_temp mean_rl ride_count
##   <date>          <dbl>   <dbl>      <dbl>
## 1 2020-04-01         10      36      84607
## 2 2020-05-01         18      33     199674
## 3 2020-06-01         23      33     341897
## 4 2020-07-01         26      38     548485
## 5 2020-08-01         25      30     604203
## 6 2020-09-01         19      25     497656

      Average ride length

• A more obvious high-low-high pattern is observed in casual riders on time taken travelling.
  Have peaks in the months of April ‘20 & ‘21 and a trough in December.
  Longer bike travels can be related more towards leisure activities or one-off rides to avoid traffic jams.
• A smoother pattern is observed for member riders.
  Shorter bike travels can be related to riding from home to work purpose which is much frequent and consistent.

tripdata_summary %>%
  ggplot(aes(x = s_year_month, y = mean_ride_length, color = member_casual)) +
  theme_grey(base_size = 12) +
  geom_smooth(se = FALSE, size = 1.1, show.legend = FALSE) +
  scale_x_date(date_breaks = "1 month", date_labels = "%b %y") +
  scale_color_manual(values=c("purple","gold")) +
  ylim(0, 80) +
  labs(x = "Month-Year", y = "Average Ride Length (Min)")

      Rides per day of week

• Bike usage increases as day reaches to weekend for Casual riders.
  Leisure activities outdoor can be the main contributor of this trend.
• Member rider trips is more consistent throughout the week.
  Members tend to have a firm routine in their daily schedule.
  Be it with work purpose, personal errands & leisure activities combined.

tripdata_summary_daily%>%
  ggplot(aes(x = factor(day_of_week,level = c('Monday', 'Tuesday', 'Wednesday',
                                              'Thursday', 'Friday', 'Saturday', 'Sunday')),
             y = ride_count, fill = member_casual)) +
  theme_grey(base_size = 12) +
  geom_col(width = 0.50, show.legend = FALSE) +
  facet_wrap(~member_casual) +
  scale_y_continuous(labels = scales::comma) +
  scale_fill_manual(values=c("purple", "gold")) +
  labs(x = "Day of Week", y = "Ride Count") +
  scale_x_discrete(labels = c('Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun'))

head(tripdata_summary_daily)

## # A tibble: 6 x 3
##   member_casual day_of_week ride_count
##   <chr>         <chr>            <dbl>
## 1 casual        Sunday          518410
## 2 casual        Monday          292885
## 3 casual        Tuesday         281572
## 4 casual        Wednesday       286462
## 5 casual        Thursday        302551
## 6 casual        Friday          391052

      Rides per time of day

• The overall trend of bike usage are similar.
  Increasing from morning until afternoon and decreases in the evening.
  Peak usage is between 3PM to 5PM.
• The differences are the following:
  There’s an outlier observation that there are more Casual riders who rides at the midnight.
  Member riders are prevailing bike users over Casual riders.

tripdata_time_of_day%>%
  ggplot(aes(x = started_at_time, y = ride_count, color = member_casual)) +
  theme_grey(base_size = 12) +
  geom_smooth(se = FALSE, size = 1.1, show.legend = FALSE) +
  scale_y_continuous(labels = scales::comma, breaks=seq(0,140000,20000)) +
  scale_color_manual(values=c("purple", "gold")) +
  labs(x = "Time of Day", y = "Ride Count") +
  scale_x_time(breaks = scales::breaks_width("2 hours"),
               labels = label_time(format = "%H:%M", tz = "UTC"))

head(tripdata_time_of_day)

## # A tibble: 6 x 3
##   member_casual started_at_time ride_count
##   <chr>         <time>               <dbl>
## 1 casual        00:00                14507
## 2 casual        00:30                25044
## 3 casual        01:00                20887
## 4 casual        01:30                17087
## 5 casual        02:00                14243
## 6 casual        02:30                10399

      Stations traveled

• There are 763 bike stations in total around Chicago, Illinois.
• Around 9 bike stations are unused for pick-up & drop-off from both rider types.
• We can re-optimize the bike allocations and transfer those bikes to stations with frequent rider usage to solve any “lacking of bike” issue.

station_names %>% 
  summarise(total_stations = n_distinct(start_station_name))

## # A tibble: 1 x 1
##   total_stations
##            <int>
## 1            763

tripdata_stations %>%   
ggplot(aes(x = factor(stations, level = c('start_station', 'end_station')), 
             y = stations_covered, fill = member_casual)) +
  theme_grey(base_size = 12) +
  theme(axis.title.x=element_blank()) +
  geom_col(width = 0.50, position = "dodge", show.legend = FALSE) +
  ylim(0, 800) +
  scale_fill_manual(values=c("purple", "gold")) +
  ylab("Stations Covered") +
  scale_x_discrete(labels = c('Start Station', 'End Station')) +
  geom_text(aes(label = stations_covered),
            position = position_dodge(width = 0.5), vjust = 2, size = 3.5)

head(station_names)

## # A tibble: 6 x 1
##   start_station_name                 
##   <chr>                              
## 1 Eckhart Park                       
## 2 Drake Ave & Fullerton Ave          
## 3 McClurg Ct & Erie St               
## 4 California Ave & Division St       
## 5 Rush St & Hubbard St               
## 6 Mies van der Rohe Way & Chicago Ave

Key takeaways:

1. Bike usage has a seasonal trend.
   • We can prepare our marketing strategies during the ‘Fall’ and ‘Winter’ season.
   • In the colder months, we can gather unused bikes for maintenance in preparation for peak season.
   • We can also pilot test and implement software upgrades in downtrend season.
   • At the start of peak season, we can deploy marketing agents and reach-out to Casual riders for potential Member conversion sale.

2. Casual riders are more active as the day approaches weekend.
   • In a timing perspective, we can increase the pace of marketing agents reach-out during Friday, Saturday & Sunday.

3. Optimize the bike allocation and review the station usage.
   • Transfer bikes from low usage stations to in-demand stations to maximize revenue.
   • Review the schedule of bike transfers to stations.
   • We can propose to drop low usage stations to reduce cost.

Appendix:

1. Bike trips data provided by Motivate International Inc.
2. Chicago, Illinois weather data provided by Visual Crossing.

Data workflow summary:

      Data cleaning

• Merged 17 historical datasets into a single dataset.
• Checked for white-spaces that can cause analysis miss-outs.
• Ensured that all variable names are format consistent.
• Removed rows that have ‘blank’ observations.
• Removed duplicate rows in the unique variable “ride_id”.
• Removed rows with ‘zero’ and ‘negative’ minutes in the “ride_length” calculated variable.
• Removed rows that are non-commercial station trips such as: “HUBBARD ST BIKE CHECKING”, “WATSON TESTING – DIVVY”

      Data manipulation

• Created new variables to compartmentalize existing observations for grouping.
• Created subsets of the clean data to narrow analysis and plotting.
• Merged the daily weather dataset of Chicago, Illinois.

      Data archival for project closure

• Purged initial raw data.
• Retained the clean data and its subsets for future use.

Thank you. You have reached the end of content.