I'd wanted to do a vis on bike accident data for a long time. First of all, I just started riding my bike again, which made me more aware. Secondly, last summer a ghost bike appeared on a street corner a few blocks from our house. I pass this corner probably 4 - 6 times a day, and each day I'm reminded someone died on a bike there. It's haunting.
I undertook this project because I was between classes, and I wanted to work a little on my design skills within Tableau. The data comes from data.seattle.gov - open data projects are pretty plentiful these days, and Seattle publishes a lot of information. The bike data was abundant too - almost too much data for me to use on a single dashboard - but I decided to squeeze it in anyway.
Three elements of this dashboard were picked specifically to add design to the data:
- The squiggly line chart in the middle right isn't meant to give you specifics, but it is meant to show a trend. For this element, I made a standard line chart and removed both axes so it was just floating above the heat map, but aligned with the labels at the top of the heat map.
- The heat map is actually the second element that was new to me. I think it's beautiful in the spare way it shows data. Larger = more, smaller = less.
- The pie clock. This was the trickiest, and the first thing I did. In fact, how I did this determined how you read the data. It's not what you might expect, and that's because I chose this clock. More on that below.
When approaching this project, I looked around and saw a lot of people had already done this vis. Most had divided the data by weather conditions or road conditions, answering the question 'how was the weather/road when this accident occurred?' Others divided the data by month, i.e., 'how many bike accidents occur in the different months of the year?' But there was so much more specific data I didn't want to ignore - and I was fascinated by the accident rates at different hours of the day. I wondered if I could create a mechanism to allow people to look at the data through that filter - by hour, and then by day of the week or month of the year. Was the 5 o'clock hour more dangerous than 9 o'clock?
Yes, it's a little oblique I'll admit. And I couldn't get the pie clock to work AM/PM correctly - so you see all of the accidents during that hour, day and night. The map gives you the detail for the specific time. Here's how you would read this slice, for example:
- Between 10 - 11 (AM and PM), there were 102 accidents where a car hit a bike, 51 where the bike hit the car, 6 where the bike flipped, and 4 where the bike hit an object in the road or a pedestrian. You can see this by the table just below the pie clock.
- The number of accidents between 10 - 11 (again, AM and PM) were more frequent in August and September, but dropped off in October sharply. You can see this by the green line chart.
- There were no accidents between 10 - 11 (AM and PM) on Sundays in January, February, October or November. You can see this data (and those for the other days of the week) in the heat map.
- And finally, between 10 - 11 (AM and PM), it was most often clear/partly cloudy (78% of the time), Dry (83% of the time), and Daylight (75%) when these accidents did occur.
Will I go back and update this now that I've learned a thing or two or ten? Perhaps. I certainly want another crack at doing clock math in Tableau (which is hard enough in Excel as is). But I'm content to leave it right now, build some more skills, and come back to it when I can make some significant updates! Thanks for reading!