Designing Intentionally
An EV Population Analysis Backed by Data Visualization Theory
Data Understanding and Preprocessing
When I got the data set, it had more than 280,000 data points and 15 columns in CSV format, and I needed to identify some words, as they were not familiar.
VIN—vehicle identification number
County, City, State, Postal Code, Vehicle Location
Model Year, Make (Brand), Model
Electric Vehicle Type
Clean Alternative Fuel Vehicle (CAFV) Eligibility
Electric Range (miles)
DOL Vehicle ID (ID according to the Department of Licensing)
Electric Utility (electricity supplier)
2020 Census Tract (some identification number from the 2020 census)
Further understanding the data set, there were a few null values, most of which were linked to the location of the EV registered, and a few were regarding miles, so I had to remove them to maintain the quality and the consistency of the data set and the analysis.
Then I fed the data into Power BI and examined a summary of the data set via Power Query as below.
I could have used Power Query for null value cleaning, but since the data were already in Excel instead of in a database or any other format, I preferred to go with it.
Either way is possible when cleaning and preprocessing data.
While going through the data, I got the idea of what I should present in the dashboard and what I should not.
Not all the columns are meaningful to use in storytelling; for example, there are a few columns used for the identification of the vehicles, like VIN, census tract and DOL vehicle ID and also for the vehicle locations, state, city, county, postal code, latitude and longitude points.
So from all the 15 columns, I have used only 8 columns to tell the story.
During the development I have discovered that there were a significant number of 0s in the range column due to the battery range not having been researched.
Because of that, the average range is impacted, and it is showing an unusual figure in the dashboard.
So I had to filter those visuals that have used range data that contains blanks or 0s to get the average value corrected.
Other than the above, there is another filter used to select the top 10 cities based on the count of registered electric vehicles in each city, as below.
Apart from that, there are no other advanced filters, DAX measures, or complex data manipulations that have been done in the backend of the dashboard to perform this analysis.
Applying Psychological Principles
Now we come to the core part of the article. (If you haven’t read the first article, please use this link to read it, as this article is the second part of that.)
Get Ranithri Hewasiliyan’s stories in your inbox
Join Medium for free to get updates from this writer.
When deciding and arranging the visuals, I have mainly considered cognitive load with the pre-attentive attributes like color, clear shapes and sizes of the visuals.
Here I limited the dashboard to five primary visuals for the storytelling since the human working memory can only process a limited amount of information at once.
Furthermore, intentionally avoided unnecessary grid lines, labels and 3D effects and followed the narrative bias nature of the human brain in answering the below questions directly.
When the user looks at the dashboard at a glance, they can see the answers for the initial questions that trigger in their minds, such as
- Who is the market leader?
- What does the market growth look like?
- What is the percentage of the EVs in the market?
- What is the evolution of the technology of the battery?
- Which city has the most registered electric vehicles?
- Line chart for the time-series data—Connection
- Pie chart for the proportional distribution of 2-variable data—Enclosure
- scatter plot for timely quantitative data—proximity
- tree map for proportional distribution of several variables—Enclosure
And look at the shapes of those visuals and how they were arranged with the backup of Gestalt’s principles.
But in the real world, production dashboard requirements may vary due to company themes and required visuals and their positioning as per the business users.
Additionally, there are 2 KPI cards to trigger the user with the average miles an EV has driven and the total number of EVs registered in the market.
And also there are 3 slicers that can be used by users to filter the dashboard brand-wise, model-wise and model-year-wise to see the information as preferred so that users can further dive in to find more insights, which will help their decision-making process.
Likewise, the whole raw data set is presented in a way that someone can easily and quickly get an idea of the electric vehicle population of Washington State, USA.
Conclusion
When developing dashboards for real-world production requirements, it is a crucial point to follow psychology behind the data visualization to reap the full benefits of the dashboard. Since humans are using it to make decisions, it should be a human-friendly product for them to grasp the insights without feeling overwhelmed or having an unpleasant experience. In my opinion, it should be a fun and interesting task to do because the decision-making task is already overwhelming. So BI developers should always be aware of the psychology behind the data visualization to give the best output or the solution to the users.
Technical Stack
Tools—Excel, Power BI
Data Source — Washington State Department of Licensing (EV Population Data)
Techniques—Visual-Level Filters, Gestalt-Based UI Design
