3 ms·

> you can compare populations and find statistically significant differences. No, you can't. There are far too many variables involved, people confuse correlat

> you can compare populations and find statistically significant differences. No, you can't. There are far too many variables involved, people confuse correlation with causation, ignore the null hypothesis, and most of all: the people doing these calculations in the first place (on either side) have an axe to grind.

defrost 3y ago
The people doing these studies have access to the same data, access to multiple bomb detonation exposure data, and have a broad background in terms of axes to grind. The bulk of those involved want to know the true risks and aren't motivated to under or over report. Discrepencies in results stand out. Here's one study (third in a series, IIRC) on the two bombs in Japan. https://bioone.org/journals/radiation-research/volume-187/issue-5/RR14492.1/Solid-Cancer-Incidence-among-the-Life-Span-Study-of-Atomic/10.1667/RR14492.1.full https://bioone.org/journals/radiation-research/volume-187/is... You can be specific in your objections, quote the parts you take issue with, etc.
GildedIgel 3y ago
Maybe _you_ can’t but this is well studied and robust. Do you also disagree with excess death calculations from the pandemic?
- pfannkuchen 3y ago
  I actually did find the excess death calculations to be questionable. The formula for expected deaths I saw used in the media was “the average of the last 5 years of deaths”. That seems like an incredible oversimplification. Population growth and age demographic trends alone would cause the actual expected deaths to trend upwards over time, and not capturing that would make this formula underestimate expected deaths (and therefore overestimate excess deaths). I did not see this problem addressed anywhere, did anyone else?
  - gattilorenz 3y ago
    That very much depends on the source. Look at our world in data’s formula [1], and you will see it does take into account that 2020 won’t have the same death as 2019, by using a linear model with previous years. Sure, there might be a sudden and expected increase in population growth (and especially in the 70+ people), but one would have to have to explain how that happened… [1] https://ourworldindata.org/excess-mortality-covid#what-is-excess-mortality https://ourworldindata.org/excess-mortality-covid#what-is-ex...
  - GildedIgel 3y ago
    It’s an oversimplification in the media, this is a time series analysis esp. at the macro level. Then you do some fancy math and estimate the counter factual -> the delta between observed and estimated is your excess mortality. Google has a tool for doing this called CausalImpact (available as an R & Python library if you want to play with a tool that can do this analysis).
    - HWR_14 3y ago
      Thanks for the tool recommendation.
arghwhat 3y ago
Everyone has an axe to grind. In this case the "axe" is trying to avoid the lung cancer, reduced immune system efficacy, reduced cognitive development of air pollution, as well as the environmental damage (which resulting forest fires escalate) from ever increasing temperatures. Of all the axes needing grinding, this one should get serviced first. > ... blah blah imperfect data Sure, the data is imprecise but we are all very sure that "air pollution == very bad". What's your suggestion, doing nothing and letting the problem grow just because there might be an percentage error margin in the numbers for how badly we are screwed? Doesn't matter if the real numbers are a quarter or 10 times higher, it's horrible compared to the alternatives regardless - unless you are somehow arguing that solar and wind are unknowingly extremely deadly. (Nuclear is a trickier discussion - the reason it is seen as safer per Wh than even wind despite the horrible incidents is due to the massive power generated by the few hundred reactors in the world, where installing hundreds of thousands of huge windmills make construction and workplace disasters more common and an insignificant cause of death.)

defrost 3y ago

The people doing these studies have access to the same data, access to multiple bomb detonation exposure data, and have a broad background in terms of axes to grind. The bulk of those involved want to know the true risks and aren't motivated to under or over report. Discrepencies in results stand out. Here's one study (third in a series, IIRC) on the two bombs in Japan. https://bioone.org/journals/radiation-research/volume-187/issue-5/RR14492.1/Solid-Cancer-Incidence-among-the-Life-Span-Study-of-Atomic/10.1667/RR14492.1.full https://bioone.org/journals/radiation-research/volume-187/is... You can be specific in your objections, quote the parts you take issue with, etc.

GildedIgel 3y ago

Maybe _you_ can’t but this is well studied and robust. Do you also disagree with excess death calculations from the pandemic?

pfannkuchen 3y ago
I actually did find the excess death calculations to be questionable. The formula for expected deaths I saw used in the media was “the average of the last 5 years of deaths”. That seems like an incredible oversimplification. Population growth and age demographic trends alone would cause the actual expected deaths to trend upwards over time, and not capturing that would make this formula underestimate expected deaths (and therefore overestimate excess deaths). I did not see this problem addressed anywhere, did anyone else?
- gattilorenz 3y ago
  That very much depends on the source. Look at our world in data’s formula [1], and you will see it does take into account that 2020 won’t have the same death as 2019, by using a linear model with previous years. Sure, there might be a sudden and expected increase in population growth (and especially in the 70+ people), but one would have to have to explain how that happened… [1] https://ourworldindata.org/excess-mortality-covid#what-is-excess-mortality https://ourworldindata.org/excess-mortality-covid#what-is-ex...
- GildedIgel 3y ago
  It’s an oversimplification in the media, this is a time series analysis esp. at the macro level. Then you do some fancy math and estimate the counter factual -> the delta between observed and estimated is your excess mortality. Google has a tool for doing this called CausalImpact (available as an R & Python library if you want to play with a tool that can do this analysis).
  - HWR_14 3y ago
    Thanks for the tool recommendation.

arghwhat 3y ago

Everyone has an axe to grind. In this case the "axe" is trying to avoid the lung cancer, reduced immune system efficacy, reduced cognitive development of air pollution, as well as the environmental damage (which resulting forest fires escalate) from ever increasing temperatures. Of all the axes needing grinding, this one should get serviced first. > ... blah blah imperfect data Sure, the data is imprecise but we are all very sure that "air pollution == very bad". What's your suggestion, doing nothing and letting the problem grow just because there might be an percentage error margin in the numbers for how badly we are screwed? Doesn't matter if the real numbers are a quarter or 10 times higher, it's horrible compared to the alternatives regardless - unless you are somehow arguing that solar and wind are unknowingly extremely deadly. (Nuclear is a trickier discussion - the reason it is seen as safer per Wh than even wind despite the horrible incidents is due to the massive power generated by the few hundred reactors in the world, where installing hundreds of thousands of huge windmills make construction and workplace disasters more common and an insignificant cause of death.)