Understanding the Spatial Patterns of YouBike 2.0 at NTU

Final Project Introduction to Econometrics, Spring 2023

Members

• Bing-Chen Chiu (邱秉辰)
  • Double major in Information Management & Economics at NTU.
  • Data preprocessing (using Python), visualization, and discuss regression models.
• Kai-Jyun Wang (王愷均)
  • Major in Economics at NTU
  • Establishment and implementation (using R) of regression models.

Motivation

Renting/returning YouBike 2.0 in NTU campus can sometimes be troubling, and the situation of bike shortage at NTU has deteriorated in 2023 (Chin-Sum Shui, 2023). In this project, we try to understand the spatial patterns of ridership of YouBike at NTU and look for determining factors of riderships within and nearby NTU.

Data Acquisition & Preprocessing

Raw data of renting records

• 臺北市公共自行車 2.0 租借紀錄

Candidate stations

• 103 stations within and nearby NTU are manually selected.

Preprocessing

• Data manipulation using Pandas 2.0.14
  • Subset the renting records that occurred on weekdays and both renting/returning station are belong to the list of selected 103 stations.
• Obtains geographical information by Google Maps API
  • latitude & longitude information
  • bicycling-distance between two given places
  • Nearest MRT station
  • Nearest dorm

The Resulting Dataset

• We split a single day into 4 time segments:

  • A: Rent time between 07:00 - 10:59
  • B: Rent time between 11:00 - 14:59
  • C: Rent time between 15:00 - 18:59
  • D: Rent time between 19:00 - 23:59

• For each station, we have:

  • The official name of this YouBike 2.0 station and its capacity
  • The latitude & longitude of the station
  • The total flow of each time segment
  • The bicycling-distances to our selected landmarks

• We also construct a table of bicycling-distances between each pair of stations.

• Remark: The difinition of flow

  A flow of station $i$ at time segment $t$ is: $$FR_i^t = \frac{RE_i^t - RT_i^t}{C_i}$$

  • $RE_i^t$ denotes the total number of \textbf{rents} from station $i$ at time segment $t$.
  • $RT_i^t$ denotes the total number of \textbf{returns} to station $i$ at time segment $t$.
  • $C_i$ denotes the capacity of station $i$.
  • $i \in {0,...,103}, \ t \in {A,B,C,D}$.

Data Visualization

We use foursquare studio as our main visualization tool

• Red dot means the station tends to "flow out"
• Blue dot means the station tends to "flow in"
• The thickness of the line represents the ridership

Time A - Tend to flow into the campus

Time B - Mixed (also less rides)

Time C - Tend to flow out of the campus

Time D - Back to the dorms

The Regression Model

Ordinary Least Squares

Consider an OLS model $$FR_{i}^t = \alpha_t+D_i:\beta_{t}+\varepsilon_{it}$$

where $D_i:\beta_t$ is the spatial fixed effect regarding the facilities nearby.

Spatial Lag Model

Consider a SLM. $$FR_{i}^t = \rho W_iFR_{t}+\alpha_t+D_i:\beta_t+\varepsilon_{it}$$ where $W_i$ is the ith row of $W$ defined above and $D_i$ are the distances from station $i$ to the predetermined landmarks.

The List of Predefined Landmarks

Results

Ordinary Least Squares

Ordinary Least Squares with $D^e = \frac{1}{1+D}$

Step AIC Results by backward selection

Time A

Time B

Time B (with SLM)

Time C

Time D

Conclusion

• If there is a good reason to consider geographic factors in the model, the spatial autocorrelation can be captured.
• Our model shows some characteristics:
  • Flow into the campus in the morning and flow out in the afternoon (MRT, conversely)
  • Flow into the dorm area shows insignificant results.