Citibike system data reference // steve gattuso

I’ve done a good amount of work with Citibike trip data and have been frustrated quite a few times along the way with some of the quirks of the dataset and lack of documentation. Information about the dataset seems to be spread thin across a lot of blog posts from data science students doing school projects (myself included, hehe), making it hard to find an abridged summary of what to expect going into it.

This page is my attempt at providing such a reference. I hope it saves you time and grief!

Trip history

Data dictionary

The Citibike data page claims that there are two schemas for the datasets but in reality there are three. The v2 format was briefly in use and was basically the v1 format with capitalized column header names.

v1: June '13 → Feb '21 (except Oct '16 → Mar '17)

tripduration
starttime
stoptime
start station id
start station name
start station latitude
start station longitude
end station id
end station name
end station latitude
end station longitude
bikeid
usertype
birth year
gender

v2: Oct '16 → March '17

This one is kind of an oddball that has the same columns as the v1 format but with capital letters.

Trip Duration
Start Time
Stop Time
Start Station ID
Start Station Name
Start Station Latitude
Start Station Longitude
End Station ID
End Station Name
End Station Latitude
End Station Longitude
Bike ID
User Type
Birth Year
Gender

v3 Feb '21 → Present

ride_id
rideable_type
started_at
ended_at
start_station_name
start_station_id
end_station_name
end_station_id
start_lat
start_lng
end_lat
end_lng
member_casual

Things to keep in mind

started_at end ended_at times are in the local America/New_York timezone.
The dataset does not differentiate between EST and EDT during the November daylight savings time switches.¹
- My solution to this is admittedly a punt- just throwing away any ambiguous timestamps and calling it a day.
The smaller JC- prefixed files are for Jersey City² trip data.
The start and end station identifiers can be null.
- I haven’t yet discovered what a NULL value means but I suspect it has something to do with a station being taken offline?

This paper also mentions this. ↩
See this email thread. ↩