I’m heading to Chicago for Thanksgiving and for most of December and while the temperature may not be conducive to biking around the city, I wanted to check out some data Chicago makes available on their bike-share program, Divvy.
It’s an extremely rich dataset. Although Chicago’s Open Data Portal only covers 2013-2019, there’s over 21 million records in the data, along with some spatial data and basic demographic data for Divvy subscribers. Luckily, the actual Divvy site has data from January 2020 all the way up through October 2023, so I’ll be able to continue some analysis and forecasting into the current day.
Forecasting is something that I’m starting to practicing more of. I’ve done some basic forecasting/predictions for some Gaelic Football Inter-County games but that has been fairly simple – simply taking previous results and scores to create offensive/defensive rating scores and then running head-to-head comparisons through a simulation model. With more practice and data, I can start running more complex forecasting and prediction models on everything from GAA games to my Vision Zero data.
With that being said, it’s time to dive into Divvy’s data!
Unfortunately, the Divvy data doesn’t have a unique identifier for users of the system. This means that both subscribers and one-time customers will be counted as a unique rider every time they hop on a Divvy bike. That’s not ideal, but can still give us an idea of who’s using bike-share. The data also has the exact start and stop time a bike was checked out and returned, as well as the station where the ride started and the station where the ride ended.
Subscriber’s of Divvy’s system made up 75% of all trips from 2013-2019. Of those subscriber riders, 74% were male and 26% were female. Those are some dedicated Divvy-riders and speaks to the viability of Divvy’s network and the ability to bike to different destinations throughout the Divvy network.
We can also see the Divvy stations that have the most amount of rides started and ended from below.
| Station Name | # Of Rides Started |
| Streeter Dr & Grand Ave | 322,534 |
| Lake Shore Dr & Monroe St | 296,230 |
| Clinton St & Washington Blvd | 285,137 |
| Canal St & Adams St | 274760 |
| Theater on the Lake | 248,945 |
| Station Name | # of Rides Ended |
| Streeter Dr & Grand Ave | 365,462 |
| Clinton St & Washington Blvd | 280,744 |
| Lake Shore Dr & Monroe St | 273,965 |
| Theater on the Lake | 267,808 |
| Canal St & Adams | 264,702 |
Both tables have the exact same stations, indicating that many riders start their rides from these locations but also many riders end their rides at these locations.
Divvy riders definitely skew younger. The vast majority of riders are between the ages of 20 and 40, with most of those in the 25-30 year old grouping. Props to the not insignificant amount of Divvy riders over 60, though!
Breaking member ridership down by gender shows that there’s not much change in the age of riders.
Enough with some exploratory data analysis…let’s look at some trends!
Divvy Ridership Trends
I’m a bike person but I’m not the Lycra-clad kind of bike person. I’ll wear workout clothes when I ride my bike sometimes, sure. But what I’m most interested in is how biking can be an alternative to the car for getting around. This has me interested specifically in weekday trends of Divvy ridership and how prevalent ridership is during work-day commute times.
While driving and public transit remain the largest transportation mode share in Chicago, the amount of people biking in the city has increased the last two decades. Chicago is pretty dense (much denser than San Antonio, for example), making the city generally conducive to both walking and biking. While there’s certainly more work to do in making biking safer, biking is a popular option for commuting if you’re able to do it.
The chart below aggregates all weekday Divvy rides in the data, making no distinction between years. There’s two clear spikes in ridership – one right around 8:00 AM and one right around 5:00 PM. This indicates that a lot of people are using Divvy during prime work-time commuting hours. What’s also interesting is that there’s a lot more rides during the post-work peak in the afternoon than the pre-work peak in the morning. Maybe some people who took the train or bus to work in the morning are opting for a Divvy ride in the afternoon. Or they’re taking Divvy to a “third spot” for drinks with co-workers before heading home.
Weekday ridership trends during the “Pandemic Era” show the same work commuting pattern as the pre-Covid chart, but ridership starts to tick upwards beginning in 2020. More people were likely looking for socially distant ways to hang out with friends or break up the monotony of working from home by getting out on a Divvy bike.
If we step back and look at yearly ridership aggregate totals, there’s an encouraging upward trend from Divvy’s start in 2013 through 2019, nearly hitting 4 million rides in 2017. There’s a slight drop in 2018 before a rebound in 2019. There’s another slight dip in 2020 but ridership absolutely explodes beginning in 2021, increasing by 58%. There’s even more riders in 2022. It remains to be seen if 2023 ridership can match ’21 or ’22 levels, but Divvy usage is still much higher than pre-pandemic levels. It’s clearly a popular option for getting around Chicago.
The heatmap below shows daily Divvy rides for 2021, when the system saw its huge jump in ridership. The darker the red means that more Divvy bikes were used on those days where the red is a lighter color. Each day of each individual month is evaluated against each other, so a dark red square in January doesn’t mean that ridership was at the same level as a dark red square in July.
January 2021 saw some pretty high ridership and if you’ve ever been in Chicago during January, you know that the weather doesn’t mess around. Ridership fell off in February, which is typically the worst winter month in Chicago, but many of the days still saw a few thousand rides. Ridership really took off in the summer months. June, July, and August all saw multiple days over 35,000 Divvy rides and September 2021 had 4 days over 30,000 rides.
Where are riders going?
We know that Divvy ridership has exploded since 2020 and has been on a steady upward trend since the program began in 2013. But where are all these riders going? Let’s take a look at the most popular station-to-station routes and examine how far Divvy users are typically riding.
The most popular Divvy route in the data is the Lake Shore Dr & Monroe St to Streeter Dr & Grand Ave. Streeter Dr & Grand Ave station shows up several times in the top 10 as a starting point for Divvy trips.
Many of these routes are along Lake Michigan, where the Lakefront Trail hugs the water. It’s a beautiful route, following Lake Michigan going as far north as Loyola University and further south through the Museum campus and Soldier Field. I highly recommend grabbing a Divvy and riding along the Lakefront if you’re in Chicago during the warm summer months.
Forecasting Divvy Ridership
Ok, we know that Divvy is popular and a lot of people use the bike-share system in Chicago. Let’s see what forecasting can tell us about the next 12 months or so of Divvy ridership.
I’m going to forecast from November 2023 through September 2024, so almost a full year of ridership. Then I can come back and re-visit my forecast at the beginning of December and evaluate my November 2023 forecast.
Using a SARIMA (Seasonal Autoregressive Integrated Moving Average) model, I’m going to do an in-sample prediction to start and then see what the forecast says about the next 12 months. An in-sample prediction is testing the model against data that’s already been recording, essentially testing how well the model is predicting known outcomes.
I chose to use a SARIMA model because it takes into account the seasonality of the data. Divvy rides are going to be affected by the weather. Not as many people will want or choose to ride a bike during January in Chicago as will hop on the bike during May or June. This means that the weather and the time of year will have a major impact on ridership.
The model is pretty close to the historical data. It over-predicts ridership in the first few years of Divvy being on the streets before coming very close in 2017 and 2018. Actual ridership over-performed the predicted amount in the summer of 2019 and 2020, but the model over-predicted the last 3 summers.
If we look at what the model is forecasting for the next 12 months, we can see that the model expects the same season trend to occur. Divvy rides are already decreasing as the weather gets colder and then will start to climb once spring rolls around and the weather warms up. The forecast calls for ridership to peak in July 2024 before slowing down again.
What It All Means…
Bike-share is strong in Chicago and shows no signs of slowing down (except in winter). Ridership has increased since Divvy’s inception, despite a small downturn this summer from the previous highs of 2021 and 2022. As the system continues to expand and add more stations throughout all 50 wards of Chicago, I expect more and more people will continue to use Divvy. Combined with a concerted effort to create a safe bicycle network, Chicago is on-track to become a world-class bicycling city.
Now, please no more NASCAR in downtown Chicago!
Where to Find the Data and Analysis:
If you want to download the data for yourself and check out the code I used to perform the analysis, take a look around my GitHub and Chicago’s Open Data Portal and the Divvy site. Feel free to ask any questions or leave any comments or ideas for future projects!
Leave a Reply