Read on Twitter
Social media, statistics, and fake news. A quick thread:
Somebody in my social media stream posted a gushing review that "Sweden has reached herd immunity, because COVID cases are falling steeply" (see attached picture).
Somebody in my social media stream posted a gushing review that "Sweden has reached herd immunity, because COVID cases are falling steeply" (see attached picture).
That indeed is a steep decline at the end - impressive. Let& #39;s zoom in a bit. The curve is plotting a 7-day rolling average. The decline there looks *drastic*.
In fact, it looks so drastic that it arose a suspicion: Is the data OK? 7-day-rolling averages don& #39;t usually take that sort of steep nosedive. You& #39;d need essentially zero cases to get that sort of steepness. Let& #39;s check the actual underlying data:
This is rather odd. Why would we drop from ~20 new cases per million people to ... zero?
Something is not right. According to my own blog post ( http://addxorrol.blogspot.com/2020/05/my-self-help-guide-to-making-sense-of.html)">https://addxorrol.blogspot.com/2020/05/m... I advise to "seek out primary sources".
Something is not right. According to my own blog post ( http://addxorrol.blogspot.com/2020/05/my-self-help-guide-to-making-sense-of.html)">https://addxorrol.blogspot.com/2020/05/m... I advise to "seek out primary sources".
It turns out that finding the *official* swedish number of new cases per day is a bit difficult, at least none of the major search engines point to the primary source.
Took me 15 minutes to find https://www.folkhalsomyndigheten.se/smittskydd-beredskap/utbrott/aktuella-utbrott/covid-19/statistik-och-analyser/bekraftade-fall-i-sverige/">https://www.folkhalsomyndigheten.se/smittskyd... which links to https://experience.arcgis.com/experience/09f821667ce64bf7be6f9f87457ed9aa">https://experience.arcgis.com/experienc... which actually ...
Took me 15 minutes to find https://www.folkhalsomyndigheten.se/smittskydd-beredskap/utbrott/aktuella-utbrott/covid-19/statistik-och-analyser/bekraftade-fall-i-sverige/">https://www.folkhalsomyndigheten.se/smittskyd... which links to https://experience.arcgis.com/experience/09f821667ce64bf7be6f9f87457ed9aa">https://experience.arcgis.com/experienc... which actually ...
... has the data. Apparently, there is no miraculous drop to zero.
Playing around with the ourworldindata interface, I finally manage to plot the raw data. What is this? Data missing?
Anyhow, summary: (1) Secondary sources have a data ingestion problem which leads to (2) calculating a wrong statistic (0 cases per million people) which leads to (3) a wrong curve (steeply descending rolling 7-day average) which leads to (4) people on Facebook claiming Sweden ...
... has reached "herd immunity". And searching for the "right" data actually takes 15+ minutes.
I think we have a structural problem.
I think we have a structural problem.
