In the data from Europe Sweden is an exception, as article explains:
"In the data that we present on Our World in Data, which comes directly from the European CDC, deaths in Sweden are shown by date of death, while deaths in other countries are shown by date of report."
"In practice this means that Sweden might today only report 10 deaths for yesterday, but once reporting is complete the death count for that day might increase to 40."
If you plot by reporting date, you're gonna get a death graph that has approximately the same shape as the true data, but it's going to be delayed by a week or two.
If you plot by actual death date, you're gonna get a death graph that's completely accurate for historical values, and constantly underestimated for the last week or two.
In the first case, trend curves will always be correct, but you should have acted on them two weeks ago. In the second case, trend curves will always be wrong, so you have to apply a forecast before you act on them.
Both presentations have errors, both require you to think before you act. But you might be right that plotting by reporting date creates an error that doesn't screw up regular people's intuitions too badly.
Either way, statistics is hard and unintuitive, and presentation of statistical data has so many pitfalls and ways to intentionally or unintentionally mislead people's thinking. :-/
7 comments
[ 4.8 ms ] story [ 29.3 ms ] thread"In the data that we present on Our World in Data, which comes directly from the European CDC, deaths in Sweden are shown by date of death, while deaths in other countries are shown by date of report."
"In practice this means that Sweden might today only report 10 deaths for yesterday, but once reporting is complete the death count for that day might increase to 40."
When tje late arriving fact is sales or orders shipped, it causes a ton of heartburn.
When this metric is expenditures or customer complaints, it gets a shrug. I frequently hear jokes about hoping the fact "can be as late as it wants."
It's fundamentally an exploitation of human's loss aversion bias.
When it involves public health, it feels particularly irresponsible.
If you plot by actual death date, you're gonna get a death graph that's completely accurate for historical values, and constantly underestimated for the last week or two.
In the first case, trend curves will always be correct, but you should have acted on them two weeks ago. In the second case, trend curves will always be wrong, so you have to apply a forecast before you act on them.
Both presentations have errors, both require you to think before you act. But you might be right that plotting by reporting date creates an error that doesn't screw up regular people's intuitions too badly.
Either way, statistics is hard and unintuitive, and presentation of statistical data has so many pitfalls and ways to intentionally or unintentionally mislead people's thinking. :-/