Explore historical bike trip data to identify key metrics and trends to drive improvements to the company’s marketing strategy.
Consider a fictitious bike-share company in Chicago. The director of marketing believes the company’s future success depends on maximizing the number of annual memberships. Therefore, it is required to understand how casual riders and annual members use the company’s bikes differently. From these insights, a new marketing strategy can be designed to convert casual riders into annual members. Company executives must approve any recommendations, so they must be backed up with compelling data insights and professional data visualizations.
We started by downloading the public data sets from the AWS S3 bucket. We choose the latest complete year of data, which is from January 2022 (202201) to December 2022 (202212). Since each data set is a zip file, we created a script to scrape and unzip the csv files to our project’s data folder.
The next step is to read in all of the CSV files inside the directory
and build one large data set by merging each of the smaller data sets
together as diagrammed below. 
Since we have a particularly large number of files, a loop can take some time. Therefore, it is advantageous to use the fread() function from the data.table package. This function is highly optimized and is much quicker than other functions for reading in large data files.
At this point, our combined data set consists of 13 columns of 5,667,717 rows. The data format is long, so we can now start adding columns representing calculated values of interest to our analysis.
There are a few problems we will need to fix:
-
We will want to add a calculated field “ride_time” to the entire data frame for analysis of how long each bike-share trip lasted.
-
There are a few lines where started and ended times seem suspiciously close to each other. These “non-rides” should be excluded from analysis by removing all trips that are less than a minute.
-
The data can only be aggregated at the ride-level, which is too granular. We will want to add some additional columns of data – such as day, month, year – that provide additional opportunities to aggregate the data.
Now we can start doing our exploratory analysis. We start with a few aggregations of mean and median ride_times split across the member vs casual categories. Then we generate a few plots to visualize the differences between member and casual usage.
Finally, we export a few of our aggregate calculations to CSV files for further analysis or presentations.
The first thing we notice from the preliminary analysis is that casuals tend to have longer trips than members:
member_casual ride_time (min)
casual 29.75593
member 12.99177
This would imply that members rent bikes for commuting or other focused intentions like commuting or errands whereas casuals rent bikes for leisure or scenic rides.
The second thing we notice is that casual trips on weekends are much longer than their weekday counterparts. Whereas member trips are more consistent weekend or weekday.
day_of_week member_casual ride_time (min)
1 casual 34.77775
2 casual 29.78813
3 casual 26.36376
4 casual 25.25931
5 casual 26.07710
6 casual 28.63388
7 casual 33.30770
1 member 14.35764
2 member 12.53066
3 member 12.38177
4 member 12.36430
5 member 12.55655
6 member 12.80298
7 member 14.47336
When calculate the number_of_rides each type of user books on average and plot that against the day_of_week, we see that members and casuals have similar bookings on the weekend. However, members book many more rides during the weekdays than casuals.
This hints that incentives for commuting passes or other weekday promotions may entice casuals to become members.
NOTE: Execute this script only for first call to initialize environment
This list of commands in located in the 0-config.R file.
install.packages("archive")
install.packages("lubridate")
install.packages("data.table")
install.packages("tidyverse")
install.packages("ggplot2")