Debunked: J. Marvin Herndon's "Geoengineering" Articles in Current Science (India) and IJERPH

I think this thread needs a debunk summary in the OP, as most of the debunking happens in the course of the discussion; and the current discussion needs to be translated for the non-statistically inclined as to how it affects Herndon's claim.

Yeah, It's still ongoing though. Don't want to jump the gun with something that needs retracting.
 
I think this thread needs a debunk summary in the OP, as most of the debunking happens in the course of the discussion; and the current discussion needs to be translated for the non-statistically inclined as to how it affects Herndon's claim.

Thanks. My head was starting to spin.
 
So I've updated the spreadsheet with a Comparison of Table 4's major elements with the actual calculated masses of the elements in the un-leached fly ash.

The first thing is still the units are incredibly off, but even if you shoehorn it into a percentage the actual values don't make sense either. He seems to have correct figures for Silicon (22.7%), but then Aluminum is 7, when it should be 14, Calcium 4.03 when it should be 3.78

It's incredibly sloppy, but difficult to explain how sloppy it is.
 
Elemental mass percentage is calculated by figuring the percentage of the element, based on the the molar mass of the element (and oxygen) and the number of atoms. Like:

http://www.webqc.org/molecular-weight-of-SiO2.html

This percentage is then multiple by the percentage of the oxide in the sample to get the percentage of the element in the sample.
 
I think this thread needs a debunk summary in the OP, as most of the debunking happens in the course of the discussion; and the current discussion needs to be translated for the non-statistically inclined as to how it affects Herndon's claim.
Take a guess how long it took me to figure out what a "logarhythm scale" was :) these two sites made it easier for me.. if maybe someone wants to try to explain that aspect in a seperate 'general discussion' thread that can be linked too ??



http://www.explainxkcd.com/wiki/index.php?title=1162:_Log_Scale
http://mathbench.umd.edu/modules/misc_scaling/page09.htm


This
rating low probabilities by the number of 'nines' in the decimal expansion of the probability of their not happening: for example, a system which will fail with a probability of 10−5is 99.999% reliable: "five nines" https://en.wikipedia.org/wiki/Logarithmic_scale
Content from External Source
i cant comprehend at all.

means and medians and ratios.. :eek:
 
Another thing, his justification for normalizing to aluminum:

The average elemental composition of each of the 38 elements from the 23 different sources of
European coal fly ash leach studied by Moreno et al. [10], presented as ratios relative to aluminum,
is shown in Figure 2 as a function of Atomic Number. Normalization to one common element, in this
case aluminum, makes comparisons possible when total mass or total volume is not available.
In this
plot, the less abundant leachate element ratios are not shown. Note that aluminum (Atomic Number 13),
strontium (38), and barium (56), elements which are sometimes determined in post-spraying rainwater,
are relatively abundant.
Content from External Source
Seems specious, as the figures are already in ratio form. All this does is force the aluminum figures to be exactly the same. Normalizing to aluminum would only make sense if it were mixed in with some different elements you wanted to exclude from the total mass of the sample.
 
Funny how the peer-reviewer didn't ask for the rainwater data...
This paper could not have passed substantive peer review by anyone acquainted with environmental forensics.

With data unpublished the paper could not have been reviewed and without experimental details the experiment cannot be reproduced, then he also failed to provide a control...... All of that is a requirement for publication.

The Publisher states:

Full experimental details must be provided so that the results can be reproduced by other groups. IJERPH encourages authors to publish all experimental controls and full datasets as supplementary files (please read the guidelines about Supplementary Materials carefully and references to unpublished data).
http://www.mdpi.com/journal/ijerph/instructions
Content from External Source
 
Does anyone know what he is referring to exactly by "At a 99% confidence interval, they have identical means (T-test) and identical variances (F-test);" What figures is he using to calculate this?
 
Seems specious

It's a perfectly cromulent method of determining the percentage of how much of a sample came from Source 1 and how much came from Source 2. Determining how much plutonium leaked from Fukushima, for example, is confounded by the fallout from nuclear weapons testing. But each source has certain combinations of radioactivity ratios or mass ratios that make determining the percentages relatively easy. For example, the radioactivity ratio of Pu-238 to Pu-239+240 from Fukushima is 2.5; but from global fallout of nuclear weapons it is 0.026. If a sample has that ratio at 2.0 then the fraction that came from Fukushima is (2.0-.026)/(2.5-.026) = 80%

Herndon didn't approach it from that angle though. For him it was a black or white hypothesis.
 
Does anyone know what he is referring to exactly by "At a 99% confidence interval, they have identical means (T-test) and identical variances (F-test);" What figures is he using to calculate this?
T-test and F-test are used to compare two samples. So I guess he applied the tests to each element/aluminum ratio. Which is invalid because of multiple comparisons, although here none of the differences were significant (if we believe him), so it doesn't matter anyway.
The F-test is sensitive to normality, and the Moreno data are very non-normal, so the F-test is not applicable here anyway.
What I don't understand is why he included the Al/Al ratio. Should we be surprised that 1=1?
There are so many problems with this paper, any reviewer should have immediately rejected it.
 
T-test and F-test are used to compare two samples. So I guess he applied the tests to each element/aluminum ratio.
Just to clarify, it's used to compare the means and variances of samples from two populations/treatments, or in some cases, to compare those values from one sample set to a value that represents the null hypothesis. You can't do a T-test or F-test to compare two individual numbers; these are tests that use the sample variance (a measure of how "spread out" the individual measurements are in a sample) as part of the calculation. So, in this case he must have compared the means of the ratios in any such test, but then showed the ratio of the means in his figures.

Unfortunately this kind of thing is not easy to explain in a simple and clear way, with no maths. The most important problem with his analysis is that he's drawing conclusions from F- and T-tests that, by design, can't be made from those tests. But the general public can't be expected to grasp this. It's the sort of thing that peer-review is supposed to filter out.
skephu said:
There are so many problems with this paper, any reviewer should have immediately rejected it.
^This.^ If it was reviewed by qualified peers, they did an embarrassingly bad job of it.
 
I made an infographic to illustrate the top post


Easy to change, so please let me know of correction or suggestions
 
Last edited:
Well, most of these problems are minor and due to sloppiness. The major problem is the complete invalidity of the "fingerprinting" method as he didn't validate the method, i.e. he didn't show that this method can actually distinguish coal fly ash in a set of minerals of crustal origin. E.g. he should have collected a range of uncontaminated soil/rock samples (e.g. from the San Diego area) and used their composition as controls/decoys to validate the fingerprint.

Basically, in order to develop a method to recognize whether some sample of minerals is coal fly ash or not, some type of classifier should be trained and validated on a sufficiently large set of known fly-ash and non-fly-ash samples. Then if we manage to develop a reliable classifier with low false positive/false negative rates, we can apply it to the rainwater and air samples. (It is doubtful, though, that such a classifier can be built as the composition of coal fly ash is so diverse that probably it cannot be meaningfully distinguished from other crustal material; except maybe the lack of carbon may be indicative.)

Here, no classifier was trained, no negative or positive controls were used, and the match was completely arbitrarily determined. Even the applied statistical tests were inadequate and incorrectly interpreted. So the paper is 100% junk science and totally invalid.
 
Well, most of these problems are minor and due to sloppiness. The major problem is the complete invalidity of the "fingerprinting" method as he didn't validate the method, i.e. he didn't show that this method can actually distinguish coal fly ash in a set of minerals of crustal origin. E.g. he should have collected a range of uncontaminated soil/rock samples (e.g. from the San Diego area) and used their composition as controls/decoys to validate the fingerprint.

Basically, in order to develop a method to recognize whether some sample of minerals is coal fly ash or not, some type of classifier should be trained and validated on a sufficiently large set of known fly-ash and non-fly-ash samples. Then if we manage to develop a reliable classifier with low false positive/false negative rates, we can apply it to the rainwater and air samples. (It is doubtful, though, that such a classifier can be built as the composition of coal fly ash is so diverse that probably it cannot be meaningfully distinguished from other crustal material; except maybe the lack of carbon may be indicative.)

Here, no classifier was trained, no negative or positive controls were used, and the match was completely arbitrarily determined. Even the applied statistical tests were inadequate and incorrectly interpreted. So the paper is 100% junk science and totally invalid.

the detailed science/math is way beyond me, but from your post

in essence unless you know what "wrong (or right)" looks like, you can't ascertain what "right (or wrong)" looks like

just sloppy science not establishing a baseline
 
Well, most of these problems are minor and due to sloppiness. The major problem is the complete invalidity of the "fingerprinting" method as he didn't validate the method, i.e. he didn't show that this method can actually distinguish coal fly ash in a set of minerals of crustal origin. E.g. he should have collected a range of uncontaminated soil/rock samples (e.g. from the San Diego area) and used their composition as controls/decoys to validate the fingerprint.

Basically, in order to develop a method to recognize whether some sample of minerals is coal fly ash or not, some type of classifier should be trained and validated on a sufficiently large set of known fly-ash and non-fly-ash samples. Then if we manage to develop a reliable classifier with low false positive/false negative rates, we can apply it to the rainwater and air samples. (It is doubtful, though, that such a classifier can be built as the composition of coal fly ash is so diverse that probably it cannot be meaningfully distinguished from other crustal material; except maybe the lack of carbon may be indicative.)

Here, no classifier was trained, no negative or positive controls were used, and the match was completely arbitrarily determined. Even the applied statistical tests were inadequate and incorrectly interpreted. So the paper is 100% junk science and totally invalid.

It's just a fancier and more complex version of the same old false claims about water and soil testing and that things like aluminum shouldn't be found there.
 
@Mick West, can you add text describing what is shown? Such as "Fig. 6 as published by Herndon", "white data set flipped vertically", "exact match!", etc. I separated and flipped the data sets myself and I am having a hard time recognizing when it starts and what it is supposed to show.
 
I have received a reply from Weidan Zhou (whom Herndon acknowledged for professional statistics advice). He has asked me to post his following statements regarding Herndon's paper here:

1. I didn't see any raw data and didn't run any statistical analysis. Dr. Herndon described his question in the email, and I answered his question.
2. I told him that the statistical analysis he did was incorrect.

Weidan Zhou
Content from External Source
 
I have received a reply from Weidan Zhou (whom Herndon acknowledged for professional statistics advice). He has asked me to post his following statements regarding Herndon's paper here:

1. I didn't see any raw data and didn't run any statistical analysis. Dr. Herndon described his question in the email, and I answered his question.
2. I told him that the statistical analysis he did was incorrect.

Weidan Zhou
Content from External Source
That's definitely information that should be in the OP, nice one.
 
I don't like the OP as it focuses on minor issues and appears to be nitpicking. The major issue is the one I described in #179.
 
I don't like the OP as it focuses on minor issues and appears to be nitpicking. The major issue is the one I described in #179.

It's a work in progress - I intended to get a lot more stuff. The challenge here is to make it reasonably accessible, and I'm open to suggestions. I'm a bit held back by my limited understanding of statistics.

I've added the Zhou email statement, which is pretty damning - his own advisor told him that his statistical analysis was incorrect.
 
I don't like the OP as it focuses on minor issues and appears to be nitpicking. The major issue is the one I described in #179.
as a layman i cant understand post #179. i always think its great to have 'advanced' scientific analysis but i think such things also need to be explained better at a layman level. if the goal is to reach/teach the masses.

i also think there is too much info (all mixed together) in this one thread for laymen. perhaps each individual point in the OP (when its done) can have a link to a seperate thread that explains that ONE aspect better. ??

and just cause naggin is what i do best:
  • more bullets vs. block text.
  • start with CLAIM in 'ex' tags
  • even font changes may help ease complicated debunks
  • understanding increases alot once you realize what a logarythmic scale is.. which the bulk of mainstreamers dont!
 
Last edited:
as a layman i cant understand post #179. i always think its great to have 'advanced' scientific analysis but i think such things also need to be explained better at a layman level. if the goal is to reach/teach the masses.

i also think there is too much info (all mixed together) in this one thread for laymen. perhaps each individual point in the OP (when its done) can have a link to a seperate thread that explains that ONE aspect better. ??

and just cause naggin is what i do best:
  • more bullets vs. block text.
  • start with CLAIM in 'ex' tags
  • even font changes may help ease complicated debunks
  • understanding increases alot once you realize what a logarythmic scale is.. which the bulk of mainstreamers dont!
Yeah, I this is the real difficulty - explaining the problems in a way that is clear and accessible to anyone. I've passed at least a half-dozen statistics courses, and I still have to go back and look things up regularly when I put it into use. When discussing this with non-scientists, you don't want to get bogged down with the statistical nuances. But the most important, fundamental problems are pretty easy to state:

-If rainwater contains elements in ratios similar to coal ash, that doesn't mean that those elements came from coal ash; they could have just come from dust and dirt (which Herndon doesn't even consider).
-And Herndon's numbers don't even show that they're very similar, even though he clearly faked some numbers, and got other numbers wrong.
-And even with all that, he used an incorrect statistical analysis (could just give the quote from Weidan Zhou to support this at first).

You can state those key points at the top, and then start digging into evidence and details for those who are so inclined.

The plague of pay-to-publish "open access" journals is bad enough, but as I understand it, Current Science doesn't even have that excuse. I really wish an editor from that journal would join us to defend their decision to publish that paper. Or, any reviewer for either article.
 
Weidan Zhou has asked me to post this clarification here:
I don't think I could be seen as Dr. Herndon's statistical advisor. He asked me a statistical question via email several month ago. The question he asked was simplified and didn't have many details. Based on his description of the question, I told him that the statistical analysis he did was not valid. Simply to say, it was incorrect to compare two observations using t test.

I didn't know anything about the background or methods of this paper. And I had no idea that I was mentioned in the acknowledgements until yesterday.
Content from External Source
 
Back
Top