Using GIS Data To Optimize Carsharing Station Locations

July 2, 2018 by Aisling O'Reilly

Aisling O’Reilly is a Transportation Intern in our Austin office. She attends the University of Minnesota, majoring in Urban Studies with a focus on Urban Infrastructure and Environment and minoring in Geographic Information Science.

Carsharing is a rapidly growing method of transportation, particularly in large metropolitan areas where owning a car can be less appealing. However, in order for a carsharing company to be successful, they must know their potential consumers and be willing to place stations where they will get the most use. Before I started my internship at BIG RED DOG, I was able to study this phenomenon and help a carsharing company refine their business using geographic information science (GIS).

With two other undergraduate students at the University of Minnesota, we partnered with a Minneapolis carsharing company to try and understand why some of their locations were performing significantly better than others, even those in close proximity. Based on our findings, we were able to suggest locations for additional stations.

Car Sharing Carsharing
photo credit: Andreas Komodromos (cc)

We decided to analyze demographic factors of populations surrounding stations, believing there could potentially be a correlation between one or more of those factors and performance. We obtained location data of the car stations, station performance in both miles driven and revenue earned, and adjacent demographic factors from American Community Survey (ACS) data.

To begin our analysis, we obtained GIS files of census block groups in Minneapolis and St. Paul and joined these files to cleaned ACS data of demographic factors we thought could potentially influence carsharing performance (sex, race, median income, education level, marital status, age, poverty level, unemployment rate, car ownership, household size, and population). We then performed an initial visual analysis comparing each variable (as a ratio of total population) to locations of stations and their performance.

From this analysis, we were able to see a few trends right away. We then performed a linear regression to determine if the trends we were visualizing were statistically significant.

The results of our linear regression analysis showed that most variables had little or no impact on station performance, though two groups had a very strong, positive correlation to station performance: percentage of women and people aged 40 to 69. Percentage of men and median income were strongly negatively correlated to station performance.

Car Sharing Analysis
Example of visual analysis. Each dot represents a station location, size of the dot represents miles driven by cars at each location, and bar on top represents revenue gained from each location. Base map data are divided into quantiles in order to observe percentage distribution.

Based on these results, we used ArcMap’s “Select by Attributes” function to identify block groups with high concentrations of our positively correlated variables. We were able to identify two block groups in Minneapolis and two in St. Paul that we believed would be suitable for new station locations.

Linear regression results using significance level of P > 0.10

This analysis only considers a small number of variables (demographic factors); other factors such as residential density, bicycle lane infrastructure, and accessibility to public transit could be used for further study.

Written by Aisling O'Reilly

Aisling O'Reilly

Aisling O' Reilly is a Transportation Intern in our Austin office. She attends the University of Minnesota, majoring in Urban Studies with a focus on Urban Infrastructure and Environment and minoring in Geographic Information Science.