I recently stumbled on this article published by La Presse, one of Montreal’s major daily newspapers. They analyzed SPVM data on break-ins in the city of Montreal since 2015, and the resulting heatmap dataviz promised to reveal the robbery hotspots in the city.
Intrigued, I clicked on it. A giant red blob appeared above Montreal, covering the entire Plateau, from the Old Port to Ahuntsic. The neighbourhood where I live and work, the Mile-Ex, was as red as the rest. Huh.
The article mentioned that specific street corners were badly hit, so I knew that this visualization was far from conveying the richness of the data. The article also told me that the university neighbourhoods were also frequent targets, but I couldn’t tell from the heatmap whether that included campuses or whether this was purely residential.
In fact, looking at it, the heatmap didn’t tell me very much at all. Kind of a shame, I thought. An image has the potential to be a lot more insightful than a written article, but only if it’s handled with a bit more sophistication than this.
Here’s my problem with using heatmaps in this context: they generally just end up visualizing the areas that are more dense in population. It’s obvious, isn’t it? People steal more where there are more people.
The Montreal Gazette covered the story as well, linking to La Presse’s viz and claiming that in its unadulterated state, the raw data is “complicated to sift through.” Well, yes and no. Complexity and complication are not the same things. To my mind, dataviz, especially in a journalistic context, needs to offer at least a little complexity in order to add something to the conversation.
My colleague and I decided to have a look and see if we could output a better map, without coding anything, just using free online tools.
The first thing we tried was to map every infraction data point on Montreal’s map. The Montreal Data Portal indicates that for privacy reasons, the geographical locations have been aggregated to the closest street corner. Of course I understand the importance of privacy, but it does skew the data a bit, making it appear that there are more crimes in streets where the distance between two blocks happens to be smaller.
Nonetheless, it was interesting to dig into the data, find my neighborhood and all my most frequented spots in Montreal – where I get coffee, walk my dog, visit friends. Looking at the dataset in more detail, we were able to visualize an additional dimension: the time of day when the event was recorded. We used yellow for day, purple for evening and blue for night. Turns out that robberies happened about 41% of the time during the day, and almost as frequently as evenings. I am unsure whether this is actually relevant information; it’s not clear if these data mark when the event happened, or when the police was notified. Another fun fact: 18 of the Montreal break-ins happened South of Ghana, in the South Atlantic Ocean.
Obviously, Montreal’s Open Data is not perfect. Points for trying though, right?
Here’s what we ended up with.
We still wanted to get a better sense of the crime hotspots, so we used an intensity map. It only took us a few minutes, using CartoDB. As a piece of software it’s really very solid for a project like this, but keep in mind that (with the free version at least) everything you create will be published. If you want to visualize data privately, you’ll have to pay.
With our viz, the event counts are localized, giving a much more detailed picture than counting them by borough. But most importantly, the shareable URL allows users to zoom in and get details on demand.
Casting my mind back, I’ve seen plenty of interesting mappings of city crime data over the last few years. The first that comes to mind is the NYC Homicide Map by the New York Times. Back in 2012, a friend send it to me to convince me not to rent an AirBNB on the east side of Prospect Park. The data generated a fairly compelling image that ultimately convinced me that I might be safer with all the latte-mommies of Park Slope.
There was another excellent visualization using crime data created by Trulia a while back. They used an image tile overlay for the heatmap, overlaid with numbered clusters – the best of both worlds.
What I’d really like to see at this point is for someone to go a step further; to make a crime dataviz that takes population density into account. If we were to map break-ins by capita, we could really see crime prevalence much more clearly, giving the police and our communities a potentially useful tool to work with.
I guess that’s the operative word with data visualization – it needs to be useful. As a field we’re getting past the experimental spiderwebby generative-art-viz and the halcyon days of the infographic boom.
The next chapter of information design needs to be focused on interactives and images that offer insight and help us solve real problems in the real world. Beginning, perhaps, in our own neighbourhoods.
* * *
Named after two tools that help users find their way through a sea of information, Breadcrumbs & Legends is a new monthly blog that explores the spheres of dataviz, UI and design.