3 ms·

We're sitting at 220k right now so I don't see how that is such a terrible estimate when dealing with a highly multiplicative event like a pandemic. An estimate

We're sitting at 220k right now so I don't see how that is such a terrible estimate when dealing with a highly multiplicative event like a pandemic. An estimate of 10k or 1M would be something extremely off, not 170k vs 220k. You should familiarize yourself with power law processes: https://en.wikipedia.org/wiki/Power_law https://en.wikipedia.org/wiki/Power_law

IvyMike 6y ago
When he said "I am expecting it to end up under 170,000 or maybe 175,000.", if he had added "plus or minus 1.5M" that would have been clearer. Edit: In other words, when his prediction is given with an implied error of +/- 5000, that speaks to a level of confidence that as you point out, anybody familiar with power law processes would not dare to give.
- Gibbon1 6y ago
  His 'predictions' sound like an old paper I read that predicted Farr's law that AIDs would disappear in the mid 90's. Digging, found a paper from 2007 shitting on it. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1906780/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1906780/ Pubmed reference. Abstract predicts 200,000 cases total. https://pubmed.ncbi.nlm.nih.gov/2308183/ https://pubmed.ncbi.nlm.nih.gov/2308183/ I'm left with the opinion that Michael Levitt is basically a stupid person.
  - dekhn 6y ago
    ah, yes, and the part where he helped invent molecular dynamics and won a nobel prize for it was all just garbage too?
    - Gibbon1 6y ago
      Linus Pauling was pushing Vitamin C quack cures most of his life. Shockley was a believer in scientific racism and eugenics. Perhaps the proper term is fachidiot.
mturmon 6y ago
> You should familiarize yourself with power law processes... Since you brought it up: my PhD was in applied probability, and I work with heavy-tailed natural data every weekday. * As my comment noted, ML literally says 170K or maybe 175K, and flat-out dismisses 220K. The latter is an interesting figure, because we recently surpassed it. We're now well out of the 170-175K envelope and still rising at about 5K deaths per week, about 20K deaths per month [1]. Note that our weekly death toll is the implied accuracy of ML's prediction. This is a failure. If ML thought there was a power law behind the mortality from an in-process event, he should have said "170K deaths or maybe 1.7M". But he didn't say that, did he? * It's interesting to see an interview with an excellent science communicator. They will say what we know, and not a bit more than that. They will choose their words carefully. They will distinguish between speculation and accepted truth. I live in southern CA, and Lucy Jones (formerly of the USGS) is such a person, for earthquakes. These people are a treasure, because this stuff is hard and it's important to get it right if you're going to make public statements. [1] https://www.cdc.gov/coronavirus/2019-ncov/covid-data/forecasting-us.html https://www.cdc.gov/coronavirus/2019-ncov/covid-data/forecas...

IvyMike 6y ago

When he said "I am expecting it to end up under 170,000 or maybe 175,000.", if he had added "plus or minus 1.5M" that would have been clearer. Edit: In other words, when his prediction is given with an implied error of +/- 5000, that speaks to a level of confidence that as you point out, anybody familiar with power law processes would not dare to give.

Gibbon1 6y ago
His 'predictions' sound like an old paper I read that predicted Farr's law that AIDs would disappear in the mid 90's. Digging, found a paper from 2007 shitting on it. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1906780/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1906780/ Pubmed reference. Abstract predicts 200,000 cases total. https://pubmed.ncbi.nlm.nih.gov/2308183/ https://pubmed.ncbi.nlm.nih.gov/2308183/ I'm left with the opinion that Michael Levitt is basically a stupid person.
- dekhn 6y ago
  ah, yes, and the part where he helped invent molecular dynamics and won a nobel prize for it was all just garbage too?
  - Gibbon1 6y ago
    Linus Pauling was pushing Vitamin C quack cures most of his life. Shockley was a believer in scientific racism and eugenics. Perhaps the proper term is fachidiot.

mturmon 6y ago

> You should familiarize yourself with power law processes... Since you brought it up: my PhD was in applied probability, and I work with heavy-tailed natural data every weekday. * As my comment noted, ML literally says 170K or maybe 175K, and flat-out dismisses 220K. The latter is an interesting figure, because we recently surpassed it. We're now well out of the 170-175K envelope and still rising at about 5K deaths per week, about 20K deaths per month [1]. Note that our weekly death toll is the implied accuracy of ML's prediction. This is a failure. If ML thought there was a power law behind the mortality from an in-process event, he should have said "170K deaths or maybe 1.7M". But he didn't say that, did he? * It's interesting to see an interview with an excellent science communicator. They will say what we know, and not a bit more than that. They will choose their words carefully. They will distinguish between speculation and accepted truth. I live in southern CA, and Lucy Jones (formerly of the USGS) is such a person, for earthquakes. These people are a treasure, because this stuff is hard and it's important to get it right if you're going to make public statements. [1] https://www.cdc.gov/coronavirus/2019-ncov/covid-data/forecasting-us.html https://www.cdc.gov/coronavirus/2019-ncov/covid-data/forecas...