Do EV owners drive their cars?
Using Python and GIS to visually explore the Massachusetts Vehicle Census.
One in a series of projects highlighting my progress as a self-taught programmer.
I began teaching myself Python, SQL, and GIS in spring 2023 — starting from zero. I therefore welcome feedback on these projects and review for errors. And I’d be interested in taking a crack at your own data and geospatial questions too. Please get in touch.
18 July 2023
Background
One of the concerns shadowing electric vehicle uptake—and efforts to shift transportation away from fossil fuel-powered vehicles—is that EVs might be no cleaner than gas-powered vehicles when the climate impacts of manufacturing, battery production, electricity generation, and other elements of the full EV lifecycle are taken into account. Fortunately, it is now generally agreed that, in the end, EVs are indeed cleaner—even if plugged into “dirty grids”—but there is still a catch: You have to actually drive the car. In short, EV owners who leave their car in the garage have mined a lot of iron and lithium for nothing; they need to drive their car to realize the relative climate benefit over the gas alternative.
So, for a quick-take look at EV driving trends in Massachusetts, I downloaded data from the Massachusetts Vehicle Census (MVC, launched in July 2023), grabbed some U.S. Census population figures, and made quite a few calculations and joins with Python (pandas) and QGIS to produce relevant maps. As discussed below, for me, the coding challenges largely revolved around filtering and tidying csv files—including pivoting, or “widening,” data—and complex rules-based labeling and symbology expressions in QGIS.
Project outputs
On a per capita basis, EV ownership is concentrated in the western (and wealthier) suburbs of Boston, just outside Route 128 (aka I-95). Here, one finds upwards of 30 EVs per 1,000 people. Note that this figure is for “passenger” vehicles, the common non-commercial, non-government Bolts, Teslas, and Leafs that make up roughly 3 percent of all active vehicles statewide. (Note also that my analysis used point-in-time data—data as of the first day of the year.)
Do these suburbs also rack up the miles?
No! Or not as much. On average, passenger EVs are driven more miles per day in central Massachusetts than in the Boston area. Now there are likely analysis-skewing reasons for this: The high-mileage towns are in more rural parts of the state, where residents have fewer public transportation options and drivers may tend to cover greater distances no matter how their vehicle is powered. (Splicing the data by region appeared to confirm the fuel-agnostic mileage imbalance; the low mileage on Martha’s Vineyard, despite high ownership rates, likewise suggests people drive fewer miles where there’s less space to go. It’s also worth noting this MVC caveat: “Daily VMT does not show the total mileage driven in a specific geographic area, only that the actively registered vehicles garaged there produced it.”)
Another way to probe which EV owners are getting their EVs on the road is to, for each municipality, assess what percentage of all vehicle miles traveled are driven by EVs. The map looks familiar:
Fortunately for the EV cause, the western Boston suburbs, where EV ownership rates are highest, also show the highest proportion of overall miles covered by EVs. If these owners were keeping their Teslas in the garage, we might well see towns with lower EV ownership rates posting the highest proportions of EV miles traveled.
Techniques
As mentioned, creating these maps was largely an exercise in cleaning and formatting data towards creating QGIS-friendly csv files. Using the Python pandas library, I needed to sort and filter my dataframe several ways and employ the groupby method in a function I defined to isolate the years and vehicle types I was investigating. Again, the nature of the original MVC files required me to df.pivot, widen, and flatten data in order to make my life easier in QGIS. While the MVC dashboard was useful for cross-checking my work, I also wrote my own Python code to quickly query the data based on a few inputs/parameters.
When exporting dataframes to csv files bound for QGIS, I make a point of writing code that creates a logical and time-stamped filename that makes clear the data’s origin.
I performed several analyses in QGIS, trying out various visuals. In addition to many joins between vector and csv files, generating the layers I needed for the final project outputs required me to, among other steps, create centroids in polygons, write complex filter expressions, and manually mess around with graduated symbology classifications.
Data sources
In addition to the MVC data downloadable at https://geodot-homepage-massdot.hub.arcgis.com/pages/massvehiclecensus, I used shapefiles and data from:
