5 ms·

perhaps i misheard it, but i literally heard the 95% for over-65s figure this morning on the radio, which seemed to be quoting the cdc (but again, it didn't hav

perhaps i misheard it, but i literally heard the 95% for over-65s figure this morning on the radio, which seemed to be quoting the cdc (but again, it didn't have my full and undivided attention). the difference could also be first dose vs. 'fully vaccinated'.

jimkleiber 5y ago
At least this link says even for first dose it's only 91% and 87%, again, which could be off, yet seems as if it would be accurate. Again, I don't know. What I've learned this year is that statistics can be very unintuitive and I know so little about how they work, that in uncertain times like these, I almost feel reliant on choosing someone to trust who I think understands this stuff better than I.
- clairity 5y ago
  @nuerow cited nytimes[0] on the 95% estimate in another comment[1]. for the purposes of my argument, the precise number isn’t important, just that it equates to “most”. and incidentally, that’s the shape of most arguments on most things, because we’re embedded in unbounded uncertainty. the most precision we can have about most things is ‘a little’, ‘some’, ‘a bunch’, ‘most’, etc., not something like 54.736%, despite how enticing precision like that can be. it’s mostly engineering where precision like that is practical, available, and useful. [0]: https://www.nytimes.com/interactive/2020/us/covid-19-vaccine-doses.html https://www.nytimes.com/interactive/2020/us/covid-19-vaccine... [1]: https://news.ycombinator.com/item?id=28848408 https://news.ycombinator.com/item?id=28848408
  - jimkleiber 5y ago
    I think precision also matters when dealing with large numbers. If it's 60% plus/minus 30%, not such big deal if dealing with 10 people, it'd be either 3, 6, or 9 people, a range of 6 people. But if it's plus/minus 30% when dealing with a population of 100,000,000, that's a range of 60,000,000 people. That being said, yes, I agree there are many situations where precision doesn't matter that much and that qualifiers work sufficiently to describe the general sense of the situation. I think what actually drove me to write my precision rant was ironically where you stated that "easily" was hyperbole, which I took to mean as you believing it was not an accurate or precise enough description of the actions that might kill the grandma. Perhaps this thread wouldn't have happened if the person wrote "I might possibly kill grandma." I don't say that facetiously—I think these disagreements about the accuracy and precision of our observations of the world can lead to so much conflict, as it has for me in my life. So, in a way, I'm grateful for you going back and forth with me, as I've learned even more deeply that while other people may seem to use too much or too little precision to me, I may use too much or too little for them.
    - clairity 5y ago
      yes, it’s miscalibrations of risk, but not necessarily numerical imprecision, that prods my reaction. it’s possible to kill grandma, in many different ways, but it’s not probable, unless intentionally reckless or malevolent. risk is inherent to life, and we need to be continually adjusting our risk assessments so that our individual and collective behaviors approach consistency.
      - jimkleiber 5y ago
        I think the challenge with "easily" "possible" and "probable" is that they are imprecise and often very subjective. One person's "possible" could be another person's "probable." Looking at stats from Nov 2020 for infection fatality rates[1] (I didn't know the difference among infection fatality rates, case fatality rates, and crude mortality rates until after I had read this[2]), a female over the age of 80 who contracts covid-19 had a 5.759% chance of dying, or approximately a 1 in 17 chance. Is that probable or possible? I suppose it might be relative to one's normal chance of dying. Perhaps her odds of dying from breaking her hip are much higher and therefore much more probable, so covid becomes a possible cause of death. But also, that was for 80 and above, so maybe at 100 the odds are much higher than 1 in 17. However, with new monoclonal antibodies perhaps they're much lower. I think sometimes I struggle to have conversations on accuracy of risk calculations without getting more precision. Yes, I don't think it needs to go into 4 decimal places, as that precision is very hard to achieve in non-engineering settings as you had mentioned. I also think sometimes words can be way too nebulous and imprecise to bring clarity between two people, as I've done a lot of work with language, emotions, and conflict, and very often (how often is very often?), conflict comes down to different interpretations of the same word. [1]: https://www.acsh.org/news/2020/11/18/covid-infection-fatality-rates-sex-and-age-15163 https://www.acsh.org/news/2020/11/18/covid-infection-fatalit... [2]: https://ourworldindata.org/mortality-risk-covid https://ourworldindata.org/mortality-risk-covid
        clairity 5y ago
        yes, that's the calibration aspect of it. for instance, we implicitly accept a 1% lifetime chance of dying in a car accident because of the great utility of automobiles. that's a useful calibration point because it has (nearly) universal acceptance and application. a 6% chance of dying is nothing to sneeze at, but it's still far from probable, which by definition is more likely than not (>50%). and while it's 6 times death-by-car, it's also dependent on contracting covid in the first place (a prior, in bayesian terms). and that prior probability depends largely on whether the grandchild has covid themselves and how contagious they are (how much and where the virus is), and given that as an additional prior, what relevant precautions they take. this chain of dependent probabilities is also why most precautions have marginal effect (because multiplying lots of fractional numbers results in a smaller fraction). none of this chain of reasoning requires great precision by the way; a single digit of significance at each step being plenty for sound decision-making. finally, note that a 100-year-old has a 30-35% all-cause chance of death: https://www.ssa.gov/oact/STATS/table4c6.html https://www.ssa.gov/oact/STATS/table4c6.html . the addition of covid risk is not immaterial, but relatively not overwhelming.

jimkleiber 5y ago

At least this link says even for first dose it's only 91% and 87%, again, which could be off, yet seems as if it would be accurate. Again, I don't know. What I've learned this year is that statistics can be very unintuitive and I know so little about how they work, that in uncertain times like these, I almost feel reliant on choosing someone to trust who I think understands this stuff better than I.

clairity 5y ago
@nuerow cited nytimes[0] on the 95% estimate in another comment[1]. for the purposes of my argument, the precise number isn’t important, just that it equates to “most”. and incidentally, that’s the shape of most arguments on most things, because we’re embedded in unbounded uncertainty. the most precision we can have about most things is ‘a little’, ‘some’, ‘a bunch’, ‘most’, etc., not something like 54.736%, despite how enticing precision like that can be. it’s mostly engineering where precision like that is practical, available, and useful. [0]: https://www.nytimes.com/interactive/2020/us/covid-19-vaccine-doses.html https://www.nytimes.com/interactive/2020/us/covid-19-vaccine... [1]: https://news.ycombinator.com/item?id=28848408 https://news.ycombinator.com/item?id=28848408
- jimkleiber 5y ago
  I think precision also matters when dealing with large numbers. If it's 60% plus/minus 30%, not such big deal if dealing with 10 people, it'd be either 3, 6, or 9 people, a range of 6 people. But if it's plus/minus 30% when dealing with a population of 100,000,000, that's a range of 60,000,000 people. That being said, yes, I agree there are many situations where precision doesn't matter that much and that qualifiers work sufficiently to describe the general sense of the situation. I think what actually drove me to write my precision rant was ironically where you stated that "easily" was hyperbole, which I took to mean as you believing it was not an accurate or precise enough description of the actions that might kill the grandma. Perhaps this thread wouldn't have happened if the person wrote "I might possibly kill grandma." I don't say that facetiously—I think these disagreements about the accuracy and precision of our observations of the world can lead to so much conflict, as it has for me in my life. So, in a way, I'm grateful for you going back and forth with me, as I've learned even more deeply that while other people may seem to use too much or too little precision to me, I may use too much or too little for them.
  - clairity 5y ago
    yes, it’s miscalibrations of risk, but not necessarily numerical imprecision, that prods my reaction. it’s possible to kill grandma, in many different ways, but it’s not probable, unless intentionally reckless or malevolent. risk is inherent to life, and we need to be continually adjusting our risk assessments so that our individual and collective behaviors approach consistency.
    - jimkleiber 5y ago
      I think the challenge with "easily" "possible" and "probable" is that they are imprecise and often very subjective. One person's "possible" could be another person's "probable." Looking at stats from Nov 2020 for infection fatality rates[1] (I didn't know the difference among infection fatality rates, case fatality rates, and crude mortality rates until after I had read this[2]), a female over the age of 80 who contracts covid-19 had a 5.759% chance of dying, or approximately a 1 in 17 chance. Is that probable or possible? I suppose it might be relative to one's normal chance of dying. Perhaps her odds of dying from breaking her hip are much higher and therefore much more probable, so covid becomes a possible cause of death. But also, that was for 80 and above, so maybe at 100 the odds are much higher than 1 in 17. However, with new monoclonal antibodies perhaps they're much lower. I think sometimes I struggle to have conversations on accuracy of risk calculations without getting more precision. Yes, I don't think it needs to go into 4 decimal places, as that precision is very hard to achieve in non-engineering settings as you had mentioned. I also think sometimes words can be way too nebulous and imprecise to bring clarity between two people, as I've done a lot of work with language, emotions, and conflict, and very often (how often is very often?), conflict comes down to different interpretations of the same word. [1]: https://www.acsh.org/news/2020/11/18/covid-infection-fatality-rates-sex-and-age-15163 https://www.acsh.org/news/2020/11/18/covid-infection-fatalit... [2]: https://ourworldindata.org/mortality-risk-covid https://ourworldindata.org/mortality-risk-covid
      - clairity 5y ago
        yes, that's the calibration aspect of it. for instance, we implicitly accept a 1% lifetime chance of dying in a car accident because of the great utility of automobiles. that's a useful calibration point because it has (nearly) universal acceptance and application. a 6% chance of dying is nothing to sneeze at, but it's still far from probable, which by definition is more likely than not (>50%). and while it's 6 times death-by-car, it's also dependent on contracting covid in the first place (a prior, in bayesian terms). and that prior probability depends largely on whether the grandchild has covid themselves and how contagious they are (how much and where the virus is), and given that as an additional prior, what relevant precautions they take. this chain of dependent probabilities is also why most precautions have marginal effect (because multiplying lots of fractional numbers results in a smaller fraction). none of this chain of reasoning requires great precision by the way; a single digit of significance at each step being plenty for sound decision-making. finally, note that a 100-year-old has a 30-35% all-cause chance of death: https://www.ssa.gov/oact/STATS/table4c6.html https://www.ssa.gov/oact/STATS/table4c6.html . the addition of covid risk is not immaterial, but relatively not overwhelming.