Public correspondance access reveals the misleading and the cover up by GEUS.
Introduction.
This post will be in English for the benefit of foreign readers in particular Dennis Helsel.
It is written with several purposes in mind:
It will shortly review the long number of different explanations by GEUS to divert attention from the fact that proper statistical treatment of data was not performed when analysing average or median concentrations over time for a number of pesticides i.e. samples with non detects were not included.
It will show that Dennis Helsel was mislead when joining with GEUS to write at “response” to the scientific article published by Hansen et al. to demonstrate how the same data should have been properly analysed. It is my guess that by teaming up with Dennis Helsel – the very person that was referred to in the critique of not using all data – GEUS had hoped to discredit the article we wrote.
Hansen et al. artikel
It will comment on the flawed understanding of basic statistical concepts that underlies some of the “explanations” given by GEUS.
It will review the statements made by Dennis Helsel showing that no real criticism of the article by Hansen et al. can be put forward.
It wil serve as a readily available source of information for posteriority. This document is obtained through the process of public access of information legislation and it therefore belongs to the public domain.
It follows that this will be of interest only to those interested in the ways an institution lies, misleads and attempts to cover this up – and it provides the documentation for this. It will be kept short (hopefully J ). Some relevant posts allready appears on this website – they are in Danish but links will nevertheless be provided.
For more information go to http://www.straight-talk.dk/
Correspondance GEUS – Helsel.
The document obtained through public access legislation of the correspondance between GEUS and Dennis Helsel (should be read from the end and upwards by date)
Correspondance GEUS Helsel
First “explanation.
GEUS first published graphics of mean and median concentrations over time in 2011. At that time I started to critizise that these values were based only on analyses above the detection limit and I stated with reference to published articles by Dennis Helsel that all analyses should be included i.e. including the so called non-detects. It was argued that the method used by GEUS was erroneous and was giving upwards biased estimates of means and medians.
This critique was reiterated over several years without GEUS ever responding.
As the critique was made public in 2014 in an article in Maskinbladet it gained
First public critique of GEUS
some political awareness. The minister was called to a hearing 27.11.2014 in the Parliament. However when asked directly why non detects were not includede the minister could provide no answer.
Second explanation.
In a written answer to Parliament (question 271) the following explanation for not including non detects is given (my translation): … the graph would according to GEUS only show horizontal lines along the x-axis, as there are large numbers of non detects compared to detects.
Third explanation.
GEUS at a later point in time tried to argue that the reason for not including non detects was a requirement stated in the EU legislation. It was argued that monitoring should be performed in the affected part of the aquifers. This was actually part of a draft response by GEUS and Helsel but it was removed (for good reasons as such a requirement is absent from the directives).
GEUS: løgn, vildledning og statistisk inkompetence
Fourth explanation.
The last “explanation” is the misleading of Dennis Helsel. It is now postulated that samples with non detects are actualle real zero concentration i.e. really 0. It is postulated that the samples with non detects stem from groundwater that is older than the time when glyphosate and bentazone were marketed in Denmark:
“Larsen: I will infer that knowing these two pesticides have only been used for about 20 years, …Where we don’t find pesticides in the deeper part of aquifer, there is no reason to believe that the pesticide is present even in concentration below the detection limit.The travelling time to the deeper parts of the aquifers is more than 20 years.”
This statement clearly shows that GEUS is either simply uninformed or trying to deceive. Glyphosate and bentazone have been marketed in Denmark since the middle of the 1970’ies i.e. more than 40 years. This alone invalidates their argument.
GEUS report on monitoring shows a cumulative age distribution of active intakes in 2014 (figure 22) showing that only c. 20% of the groundwater is as old or older than more than 40 years corresponding to the time glyphosate and bentazone has been on the market. Thus there is a high probability that a given groundwater sample will represent water from a time at which glyphosate and bentazone was in use on the Danish market. Figure 21 additionally shows that the relationship ( the deeper the older) is actually rather weak.
Groundwater monitoring 2014
Thus there is no reason to conclude that the groundwater samples with non detects are so old as to stem from at time where glyphosate and bentazone were not on the market. It should be stated that as far as I know there exists no concrete dating of the samples with non detects. It is all speculation (as previous “explanations”) basen on erroneous knowledge of the two compounds.
What actually happened in 2011.
GEUS was asked to provide a means to follow the development in concentration in the groundwater for a number of pesticides and metabolites. They based this on a datacompilation that had been done since 1995. This consisted of all analyses split into number above the 0.1 microgram/L limit, below this limit but above the detection limit (together they constitute findings) and number below the detection limit = non detects. As concentrations were requested this was calculated based on analyses with real numbers – not knowing what to do with non detects (the work by Dennis Helsel was unknown to GEUS until recently). This is the reason that non detects were not included.
Poor statitical thinking.
A comment should be made on the attempt to argue that only the affected part of the aquifer(goundwater) shoul be included i.e. using only findings. This represents a complete backwards and inappropriate statistical thinking. If the affected part of the groundwater was of interest (the universe/population) the a sampling programme should be established for this universe. GEUS uses all samples and the ones above the detection limit (the highest concentrations) are the called the universe. This means that no proper sampling plan has been determined for the sampling of the universe and consequently no inferences can be made.
What is Dennis Helsel saying:
The following has been compiled by Jens Streibig and myself:
11/11 2015:Helsel: ”I am glad to co-author this with you as long as point 1 in my comments remains—the deletion of nondetects is not the way this type of work should be done” It becomes patently clear why GEUS had to come up with another explanation for not including non detects as Hansen et al. had clearly done it the right way.
14/10 2015:Helsel: ”I haven’t yet added a statement about what could be done with data having 10% or fewer detection. I would suggest using logistic regression to test the trend (it tests for a change in the percent that are detected, not the concentration itself…..But the lognormal (distribution) is the one traditional used in water chemistry. It is better than ignoring the nondetects, but certainly confidence intervals should also be provided, and they will be wide.” To be perfectly sure – logistic regression was performed by Hansen el al. However being a short communication this was only mentioned in the text. The demonstration of the proper method was the most important and showing what the proper estimates of mean and median concentrations were (albeit with large confidence intervals) as opposed to the misleading presentation made by GEUS.
Larsen: ”The paper by Hansen et al 2015 is mixing up the surveillance and the operational monitoring, and we will therefore claim that an estimation of values (or summary statistics) for concentration in the part of the aquifers, which has never been affected by the use of pesticides (the water is too old), is not relevant.” This is just mumbo-jumbo – it has nothing to do with surveillance or oprational monitoring to do- actually the Waterframe Directive instructs us to use both types of monitoring for trend analysis. The following text from the directive states:
“2.4.4. Identification of trends in pollutants
Member States shall use data from both surveillance and operational monitoring in the identification of long term anthropogenically induced upward trends in pollutant concentrations and the reversal of such trends. The base year or period from which trend identification is to be calculated shall be identified. The calculation of trends shall be undertaken for a body or, where appropriate, group of bodies of groundwater. Reversal of a trend shall be demonstrated statistically and the level of confidence associated with the identification stated.”
“…we know that they do not (follow lognormal distribution) since a fraction of the samples should be assigned the value zero as some of the monitoring wells collect water that is older than the date when the two pesticides were released to the Danish Market.” This has no validity as shown earlier.
7/10 2015:Helsel: ”I agree with most of what is in Hansen et al paper. Not using data that are nondetect is a very bad idea, so their argument to use censored method is quite correct… They are appropriate for estimating trends, or means and median, or the characteristic of the distribution. The methods do not estimate values for single samples”!!! Which has not been claimed by us.
”The trend line they show is perfectly fine, and is a much better representation of the trend in the data than is the line shown by using only detected values. The assumption they make is that given the detected values observed, and WITH NOT OTHER INFORMATON FROM SCIENCE, the read areas are a good model for the concentrations measured as below the detection limit” No comment needed.
6/10 2015: Helsel: ” Certainly with over 90% of data reported as nondetect, MLE is unsuitable for estimating individual chemical concentration” This has not been claimed either.
Conclusion.
Thus the overall conclusion is that GEUS has attempted to use Dennis Helsel to counter the critique put forward regarding the (lack of ) statistical treatment of monitoring data in the monitoring reports. It turns out that the article by Hansen et al. is completely valid and to the point in that it demonstrates the proper way of analysing data that are censored and shows the vast difference to the invalid method used by GEUS.
And I have not even included the misleading that took place when GEUS failed to inform about all the different “explanations” mentioned above, the confusion GEUS tried to create by trying to bring EU legislation into the picture. It is furthermore apparent that a complete blurred picture is painted of what is going on. Referring to table 19 in the monitoring report above the data compilation is on an analysis basis not boring, has no bearing on ground water bodies, findings are just the analyses with the highest concentrations and has nothing to do with ground water bodies at risk – indeed there is no consistency in the use of terms with shifting between ground water bodies at risk, part of aquifer and it seem as if GEUS equals ground water bodies at risk with boreholes which are affected….
This is all based on the document on the correspondence GEUS/Helsel. The actual response which is even more misleading but this will have to await the time when I have figured out how it can be viewed freely and therefore be able to contrast the statements made with the facts. If this takes too long I may try other avenues. In the mean time a link to the GEUS-Helsel article is given so that interested people can obtain it:
GEUS Helsel article
The last action in this story is a mail to Dennis Helsel from Jens Streibig and myself. Here we wanted to clearly explain the way in which he was mislead by GEUS. This mail is pasted below (with minor additions as a figure could not be copied).
Dear Dennis
I want to briefly return to our publication of glyphosate and bentazone monitoring results in Denmark (Attached). I fully acknowledge that the Danish situation is far too complex for foreigners when it comes to groundwater contamination with pesticides. It is a politically hot potato thanks to the way GEUS has reported concentrations without including the ND.
We were rather surprised about you being co-author of GEUS reply (Attached) to our paper. I understand from various documents we have obtained that you mostly agreed with our way to include missing values (see url: http://www.straight-talk.dk/statistics-public-access-correspondance-geusdennis-helsel/ )
We can understand you being co-author if GEUS has convinced you that ND were true zero values because according to GEUS glyphosate has only been used for the last 25 years and water samples older than that could not be exposed to glyphosate.
The very fact, however, is that both glyphosate and bentazone have been in use for the last 43 years.
GEUS provides monitoring of the age distribution of ground water (see below).
Age distribution of active water intakes (side 59, figur 22 i linket nedenfor figuren kunne ikke kopieres)
Grundvandsovervågning 2014
Thus approximately 80% of intakes samples of water were less than 40 years old.
Consequently, there is no reason to discard ND samples postulating that they represent true zero’s based on age. Indeed we question that the samples used in our analysis have been dated specifically and wonder if such documentation has been presented to you?
The criticism that too few actual values are present we accept (we did read your recommendation as summarized in table 6.11 in your book. Statistics for Censored Environmental Data Using Minitab and R. 2nd ed. Wiley.) and therefore we also did logistic regression. However, for illustration it was important to present median values with the huge uncertainty in order to put the medians presented by GEUS in a statistical context.
Finally, you should be aware that this is probably the 5th explanation by GEUS for not including all samples. Unfortunately, this is in the small language of Danish, so it is not worth to spend too much time on this for non-Danish speaking individuals.
We have followed your publications over the years and find them very valuable, albeit GEUS did not know your research when we published our Paper. GEUS clumsily try to protect their way of publishing
Best regards Claus Hansen and Jens Streibig
Jens Carl Streibig, PhD, DSc.
Professor Emeritus
Department of Plant and Environmental Sciences
Crop Sciences
University of Copenhagen
Hoejbakkegaard Allé 13
DK-2630 Taastrup
DENMARK
TEL +45 35 33 35 60
DIR +45 35 33 34 57
FAX +45 35 33 34 78
jcs@plen.ku.dk
http://plen.ku.dk/english/
http://rstats4ag.org/
Dosis sola facit venenum
Paracelsus (1494-1541)
Regrettably we have received no reply from Dennis Helsel.