Skip to content

Latest commit

 

History

History
96 lines (51 loc) · 6.89 KB

about_data.md

File metadata and controls

96 lines (51 loc) · 6.89 KB

About this site

The data here is meant to help you understand how Covid-19 is evolving in different areas of France, in one place, without bogging you down with superfluous stats

The ultimate goal is to give you more context beyond increasingly-alarming case numbers: Is that good or bad? Are we heading up or down?

To that end (in case you haven't already noticed) these plots are all interactive: zoom in on a area to get more detail, or select just a few lines by clicking the legend.

Oh, and mobile users: please turn your phone sideways.

Coming soon

  • adding alert thresholds to the indicator line plots, to help answer the inevitable question "Just how bad is it?". DONE

  • creating a map of the metropoles departments under curfew DONE

  • adding summary of departments where alert levels are going up vs down NO LONGER RELEVANT (government seems to have abandoned alert levels)

  • discussing how testing influences incidence rates IN PROGRESS

  • tracking the confinement - to see how if/when/at what rate the case counts & ICU occupancy are going down. IN PROGRESS: new bar charts are on the Overview page, plus new heatmaps of % changes by department{:target="_blank"}.

    • As this more people are working and kids are in school, this is a lighter confinement than last sprint. Accordingly progress will be slower. The big question is will it be enough to celebrate les fêtes?

About Covid-19 data

To borrow a phrase from economists, there's no such thing as a perfect dataset. Covid-19 data is the embodiment of this. It's extremely hard to establish any certainties with a novel virus that spreads so easily, manifests with such varied symptoms and outcomes, and wreaks literal havoc across so much of the world.

In such a fast-evolving environment, data collection is not going to be exhaustive, nor ideal. The New York Times, Le Monde, The Guardian and the Canadian Broadcasting Company, just to name my go-to information sources as a layperson, have all reported stories of delayed results, not enough testing of asymptomatics, technical snafus, among other problems.

This applies to all Covid-19 data, not just what is visualized here.

Nonetheless, even if Covid data is not 100% accurate, it still provides decent benchmarks of where we are, and rough signposts of where we might be heading in near future.

About this data (data caveats)

These maps display Covid-19 indicator values. The values are merely an approximation of the official government figures used to determine if a threshold has been reached.

I cannot replicate government figures completely due to the following:

  • a 3-day lag in testing numbers. Since incidence rates are based on testing data, the 'latest' incid_tous & incid_70+ values here will always from at least 3 days earlier.

  • Personnes agées for the French government means people 65 or older. However, the publicly-available, daily testing data only breaks down age in 10-year increments. As a result the maps here show incidence rates for people 70 or older( incid_70+). In other words, they are understated compared to official government figures.

  • The government examines overall incidence rates for the major metropolitan in France when classifiying its surrounding département. Until Oct 15 2020 there wasn't any data on these metropoles, so I've applied those thresholds to departments instead. This likely means the overall incidence rate in these maps (incid_tous) is also understated compared to official figures.

Data sources

The Covid-19 data comes from Santé Publique France and is updated daily at 19h15 Paris time.

The plots here are updated shortly afterwards, usually by 21h Paris time.

Other data sources:

About the data processing

When I started in April 2020, there wasn't as much data available—or it wasn't findable—so I've done additional data processing to:

  • create rolling 7-day sums for positive tests and total number of tests

  • calculate incidence rates for entire population as well as incidence rates for the 70+ age group

  • calculate percentage of ICU beds occupied by Covid patients

  • classify each département according to the alert thresholds described above...to the best of my ability with publicly-available data. See 'Data Caveats' above.

Even though rolling 7-day averages and taux d'incidence are now available pre-calculated on data.gouv.fr, I prefer to keep working with daily "raw" numbers, especially as it allows me to see testing numbers (which I hope to share more here soon).

  • The exception is metropole incidence rate data, which for now are only available pre-calculated.

You can find my data processing scripts here.

About me

Most of my professional life has been spent as a technical communicator, or herding others, at enterprise software firms in the search/analytics/business intelligence space.

Upshot: I've spent a lot of time learning new, often pretty complex, tools quickly, just to explain to techie types the ABCs of getting and analyzing data with them.

After years of this I became more interested in learning how to analyze data myself, at the command line with open-source librairies.

That's how looking at Covid-19 data and compulsively reading Covid news articles somewhat naturally became my hobby since April 2020. It was a good distraction, oddly enough, while confined to a small Parisian apartment for 2.5 months.

With the arrival of the second wave and accompanying media blitz of stats, alerts, and heatmaps, it was leaving me overwhelmed. And if someone for whom data visualization is a fun weekend activity (sometimes, at least) is overwhelmed, what about everyone else?