Hi, Jack - The key is in the error bars. For either of these datasets, the proper null hypothesis is that long-period exposure to low levels of radiation has no effect at all on cancer risk. Both datasets support that conclusion, but for the first one, any line from a +20% increase in risk over the dose range studied to a -20% protective effect would fall within all the (I would hope the standard) 95% error bars. For the later version, the fourth error bar barely creeps above zero. That's the one that looks like it falsifies LNT and comes close to suggesting a fairly high probability of a protective effect (>90% of the error bar is in negative territory, and it's relatively narrow compared to the higher-dose bars.
But I don't understand something about the error bars. The third (50-99) bar has nearly 10 times the person-years of the 4th, yet its error bar isn't that much narrower. I'm still of the mind the best conclusion from the data is no effect.
I should never have brought in the null hypothesis concept without explaining it. Mea culpa.
It's a matter of burden of proof. In any situation where we have data scatter, there's always the possibility that the observed pattern is just due to chance. That's called the null hypothesis. Anyone who proposes an explanation for the observed pattern must show that the null hypothesis is false with high probability, usually taken to be at least 0.95. In other words, the chance of coming up with the observed experimental result if the null hypothesis is true is less than 1 in 20.
In this case, the null hypothesis should be "the radiation doses had no effect". The Karunagappally data obviously does not contradict the no effect hypothesis. A Relative Risk of 1.0 is well within the 95% confidence interval for all the dose groups. Under normal rules, the no effect hypothesis must be accepted.
But if LNT is fallaciously assumed to be the null hypothesis, than the burden of proof is on the non-LNTer. She must show that, if LNT is correct, the probability of the observed experimental result is less than 1 in 20. This means as long as the sample size is small enough or the data scatter is large enough, she can't disprove LNT. Therefore, LNT must be accepted.
From concluding remarks of the Karunagappally study.\cite{amma-2021}:
"... the confidence interval obtained from the present study is much narrower than that in the previous study, suggesting a possibility that the solid cancer risk associated with the continuous exposure to low-dose rate radiation is significantly lower than that associated with acute exposure. However, such a comparison requires matching of age at exposure since the magnitude of radiation associated RRs are known to be dependent on age at exposure."
It seems as though these researchers are seeking a job for life befitting xkcd's 882 scenario. ;)
This is much like the Neel &/or Schull studies on mutational effects (next generation effects) of Hiroshima/Nagasaki survivors. The results were always expressed in terms of LNT, even though no increase was shown. Instead of saying there was no effect, they said there is no evidence of an effect larger than ??? at doses to the parents of ???. The first of these studies came out in 1956. Luckily John Boice has come out strongly saying that we should not worry about next generational effects.
The error bar at 237 mSv spans a little less than 0.2 RR. The error bar at 617 spans about 0.5 RR. I think that is close to what I would expect.
LNT isn’t science, it is religion. Unfortunately the tarot card readers are in charge and they aren’t going anywhere. Trying to convince an LNTer with data is like trying to get a cow to eat a chicken- they just aren’t interested.
"One fundamental shortcoming of the LNT model is that it does not account for natural processes that repair DNA damage. However, there is no contiguous mathematical model that estimates cancer risk for both high- and low-dose rates that incorporates what we have learned about DNA repair mechanisms and is sufficiently simple and conservative to address regulatory concerns. The author proposes a mathematical model that dramatically reduces the estimated cancer risks for low-dose rates while recognizing the linear relationship between cancer and dose at high-dose rates."
We will now wait for rebuttal from any of the numerous pro-LNT folks who have popped up in an earlier discussion on radon.
I know you don't have time to participate in these whack-a-mole debates on the Internet, but perhaps someone in the choir could step in. I could do it, but that will compromise my role as editor at Citizendium.
Take a look at what I have collected in the debate on radon (on the page linked above). It is almost ready for condensation to a summary on our Debate Guide page. That is what our journalists need - a summary from both sides, in their own words, with links to their longer arguments.
Based on your use of the figure, it appears I have not make myself clear. The key point is not that the folks that got 30 mSv over 10 years had the same cancer incidence statistically as the folds that got 600 mSv.
The key point is 600 mSv received all at once results in a significant increase in cancer but 600 mSv spread over ten years results in no detectable increase in cancer. According to LNT, the increase from 600 mSv acute is 5.7%. According to SNT that increase is 10.4%. (The SNT fit to the RERF data is higher than the LNT in this range.) According to LNT, the increase from 600 mSv spread evenly over 10 years is 5.7%. According to SNT it is 0.017%. SNT matches both the very high dose rate data and the low. LNT cannot.
Hi, Jack - The key is in the error bars. For either of these datasets, the proper null hypothesis is that long-period exposure to low levels of radiation has no effect at all on cancer risk. Both datasets support that conclusion, but for the first one, any line from a +20% increase in risk over the dose range studied to a -20% protective effect would fall within all the (I would hope the standard) 95% error bars. For the later version, the fourth error bar barely creeps above zero. That's the one that looks like it falsifies LNT and comes close to suggesting a fairly high probability of a protective effect (>90% of the error bar is in negative territory, and it's relatively narrow compared to the higher-dose bars.
But I don't understand something about the error bars. The third (50-99) bar has nearly 10 times the person-years of the 4th, yet its error bar isn't that much narrower. I'm still of the mind the best conclusion from the data is no effect.
Yes, the error bars are for the 95%, two sided, confidence interval
Could you explain this a bit simpler for those of us who are technically less gifted, including the null hypothesis bit? Thank you.
Coltrane, Reuben and others,
I should never have brought in the null hypothesis concept without explaining it. Mea culpa.
It's a matter of burden of proof. In any situation where we have data scatter, there's always the possibility that the observed pattern is just due to chance. That's called the null hypothesis. Anyone who proposes an explanation for the observed pattern must show that the null hypothesis is false with high probability, usually taken to be at least 0.95. In other words, the chance of coming up with the observed experimental result if the null hypothesis is true is less than 1 in 20.
In this case, the null hypothesis should be "the radiation doses had no effect". The Karunagappally data obviously does not contradict the no effect hypothesis. A Relative Risk of 1.0 is well within the 95% confidence interval for all the dose groups. Under normal rules, the no effect hypothesis must be accepted.
But if LNT is fallaciously assumed to be the null hypothesis, than the burden of proof is on the non-LNTer. She must show that, if LNT is correct, the probability of the observed experimental result is less than 1 in 20. This means as long as the sample size is small enough or the data scatter is large enough, she can't disprove LNT. Therefore, LNT must be accepted.
For a visual representation of Jack's explanation below, and what happens in news reports, type the following into your browser:
xkcd (dot) com (slash) 882 (slash) replacing the punctuation words with actual punctuation. substack doesn't allow links in comments.
In one image, you have an explanation for all of social science. Never, ever let a sociologist loose with a stats package.
This is the most correct answer.
Great Work! I am not a statistician, but the data you present seems true.
From concluding remarks of the Karunagappally study.\cite{amma-2021}:
"... the confidence interval obtained from the present study is much narrower than that in the previous study, suggesting a possibility that the solid cancer risk associated with the continuous exposure to low-dose rate radiation is significantly lower than that associated with acute exposure. However, such a comparison requires matching of age at exposure since the magnitude of radiation associated RRs are known to be dependent on age at exposure."
It seems as though these researchers are seeking a job for life befitting xkcd's 882 scenario. ;)
This is much like the Neel &/or Schull studies on mutational effects (next generation effects) of Hiroshima/Nagasaki survivors. The results were always expressed in terms of LNT, even though no increase was shown. Instead of saying there was no effect, they said there is no evidence of an effect larger than ??? at doses to the parents of ???. The first of these studies came out in 1956. Luckily John Boice has come out strongly saying that we should not worry about next generational effects.
The error bar at 237 mSv spans a little less than 0.2 RR. The error bar at 617 spans about 0.5 RR. I think that is close to what I would expect.
LNT isn’t science, it is religion. Unfortunately the tarot card readers are in charge and they aren’t going anywhere. Trying to convince an LNTer with data is like trying to get a cow to eat a chicken- they just aren’t interested.
Jack, I would like to include your Figure 2 in our article on Fear of Radiation https://citizendium.org/wiki/Fear_of_radiation
We publish the clearest, most concise summary of facts on each issue, with reliable sources.
Is this figure your creation, under CC-BY-SA license, or do we need go to the original source.
I searched for amma-2021 and found the earlier study behind a paywall. I also found this article, which you might be interested in: https://journals.lww.com/health-physics/abstract/2023/06000/the_relationship_between_cancer_and_radiation__a.5.aspx
From the abstract:
"One fundamental shortcoming of the LNT model is that it does not account for natural processes that repair DNA damage. However, there is no contiguous mathematical model that estimates cancer risk for both high- and low-dose rates that incorporates what we have learned about DNA repair mechanisms and is sufficiently simple and conservative to address regulatory concerns. The author proposes a mathematical model that dramatically reduces the estimated cancer risks for low-dose rates while recognizing the linear relationship between cancer and dose at high-dose rates."
Figure 2 is based on Table 5 in the Amma paper. I drew the barchart. or rather my computer did. Use it as you please.
Thanks. I have uploaded the figure to Citizendium and added it to the Talk page of our our Fear of Radiation article. https://citizendium.org/wiki/Talk:Fear_of_radiation
We will now wait for rebuttal from any of the numerous pro-LNT folks who have popped up in an earlier discussion on radon.
I know you don't have time to participate in these whack-a-mole debates on the Internet, but perhaps someone in the choir could step in. I could do it, but that will compromise my role as editor at Citizendium.
Take a look at what I have collected in the debate on radon (on the page linked above). It is almost ready for condensation to a summary on our Debate Guide page. That is what our journalists need - a summary from both sides, in their own words, with links to their longer arguments.
Based on your use of the figure, it appears I have not make myself clear. The key point is not that the folks that got 30 mSv over 10 years had the same cancer incidence statistically as the folds that got 600 mSv.
The key point is 600 mSv received all at once results in a significant increase in cancer but 600 mSv spread over ten years results in no detectable increase in cancer. According to LNT, the increase from 600 mSv acute is 5.7%. According to SNT that increase is 10.4%. (The SNT fit to the RERF data is higher than the LNT in this range.) According to LNT, the increase from 600 mSv spread evenly over 10 years is 5.7%. According to SNT it is 0.017%. SNT matches both the very high dose rate data and the low. LNT cannot.