Read on Twitter
WITTGENSTEIN'S RULER
"Unless you have confidence in the ruler’s reliability, if you use a ruler to measure a table you may also be using the table to measure the ruler." - @nntaleb, Fooled by Randomness
How does it apply to the statistics regarding the pandemic?
(thread, 1/N)
"Unless you have confidence in the ruler’s reliability, if you use a ruler to measure a table you may also be using the table to measure the ruler." - @nntaleb, Fooled by Randomness
How does it apply to the statistics regarding the pandemic?
(thread, 1/N)
2/ Example: if you read that "in Italy there were 800 confirmed COVID-19 deaths yesterday", this might give you information about
- How many people died in Italy due to COVID-19, or
- How many people got tested, or
- How reliable are Italian doctors, or
- How trustable is the gov
- How many people died in Italy due to COVID-19, or
- How many people got tested, or
- How reliable are Italian doctors, or
- How trustable is the gov
3/ Example #2: January data about COVID mortality was providing information regarding:
- COVID mortality, or
- Chinese healthcare, or
- Reliability of Chinese government, or
- Something else
- COVID mortality, or
- Chinese healthcare, or
- Reliability of Chinese government, or
- Something else
4/ Every observations has one or more free parameters and, the more they are and the more free they are, the less you know what you're measuring.
(examples below)
(examples below)
5/ In the chart below, the free parameter “who’s being tested?” in Iceland tends to “everyone”. Its incertitude decreases ("it's less free"), so the object of the observation shifts from “who’s symptomatic?” to “who has the virus?” https://twitter.com/alexandreafonso/status/1243557013759700997
6/ This is why I've been criticizing charts as the below one. We don't know what they're about.
Are they about:
- deaths?
- testing critieria?
- government reliability?
- etc?
Probably, all of the above. https://twitter.com/jburnmurdoch/status/1244368888680329218?s=20
Are they about:
- deaths?
- testing critieria?
- government reliability?
- etc?
Probably, all of the above. https://twitter.com/jburnmurdoch/status/1244368888680329218?s=20
7/ Example #4: We don't know if patients were randomized across groups.
Thus, the claimed p-value = 0.002 informs us about any of:
- the drug
- the drug administration criteria
- the doctor claiming the low p
(that said, I do believe HCQ is promising) https://twitter.com/JamesTodaroMD/status/1244385850684817409?s=20
Thus, the claimed p-value = 0.002 informs us about any of:
- the drug
- the drug administration criteria
- the doctor claiming the low p
(that said, I do believe HCQ is promising) https://twitter.com/JamesTodaroMD/status/1244385850684817409?s=20
8/ This is why, in the research below, I used "deaths regardless of cause": it's one of the most solid data points (very low parameters freedom):
- Source already exists since decades
- Not directly controlled by central government
- 0 diagnostic error http://pandemic.substack.com/p/the-elephant-in-the-room-undercounting
- Source already exists since decades
- Not directly controlled by central government
- 0 diagnostic error http://pandemic.substack.com/p/the-elephant-in-the-room-undercounting
