Comparing new Canadian COVID-19 stats between the provinces

April 21, 2020

> See the chart in action here

Coronavirus emoji
SARS-CoV-2 emoji by iXimus

The CBC publishes a bunch of great COVID-19 charts that allow you compare various rates between provinces. While they really are well-done in general, I found them lacking in one way: They all use absolute numbers rather than normalized for population (per capita). This makes it really hard to compare between provinces with vastly different populations (two orders of magnitude).

Data science as a hobby

I’m no data scientist, but I do enjoy munching numbers. Poking around the browser developer tools I found that the CBC has an API for COVID-19 data. As a bonus, they set the CORS header to allow use on any domain, which I take as implicit approval to make gentle use of it myself.

Looking specifically at daily new cases of COVID-19, I thought that it would be more informative to compare the province’s rates per capita. I settled on cases per one million people.

Pick a stack, any stack

I’ve been using Nuxt.js a lot lately, so it was a natural choice.

There isn’t a ton of benefit to server-side rendering this project. It uses Chart.js under the hood, which renders into a <canvas> element. SSR-ing the contents of a <canvas> element isn’t really possible, as there in no DOM to render.

But Nuxt has more to offer than just SSR. The additional lifecycle hooks it provides, like fetch, are ideal for making API requests. This means the API call doesn’t need to be made from the browser, hopefully resulting in a faster load time.

You can check out the github repo here.

Normalizing the data

In addition to the “new cases” API endpoint here , I also got a CSV from Statistics Canada containing the latest population data and munged it into pretty JSON with csvjson from csvkit and jq.

There is one other thing that irked me about the data. My home province, British Columbia, does not release data on Sundays. This shows as a zero datapoint for that day, which I feel is misleading. To address this, I wrote some code to map those values to null. Combined with the Chart.js options spanGaps, there is no datapoint shown on the chart for that day.

I saw the CBC take a different approach to deal with the BC’s lack of Sunday data points on one of their charts. They average the values for Saturday - Monday. I had already implemented my “span the gaps” strategy, but I’m really not sure if one is “more accurate” than the other. Drop me a line if you know.

Show me the chart, already

OK, here it is. You can compare various provinces and the nation as a whole. Just click on the regions you want to enable in the legend at the bottom. Hover/clicking on individual points on the chart will show you the absolute value.

It is just a quick prototype and it is not very mobile-friendly. The legend does not work great on mobile and we could probably exclude the data points from before about March to make it better fit the screen.

Limitations

Unfortunately, the numbers never tell the whole story. These are just the confirmed cases and each province tests differently.

Here are a few more factors to keep in mind:

  • Tests can get backed up, resulting in a backlog of cases all being reported at the same time.

  • There is no distinction between a large number of cases in a single facility or community spread.

  • In low-population provinces, a small absolute number of cases can result in a huge spike in cases per million people.

Conclusion

I find numbers comforting. It helps me feel like I know what is going on and I can plan accordingly. This may just be a comforting illusion but, either way, I will take it.

Stay safe and look out for each other.