Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closeFigure 2 does not support chloride claims
Posted by Garbarino on 06 Dec 2013 at 18:14 GMT
The authors claim that lakes with drilling sumps in their catchments had significantly higher concentrations of chloride (Cl−) than either the control lakes or lakes impacted by permafrost thaw slumps. I think there is a mistake here because Figure 2 does not support this claim. The standard deviation of drilling lakes is very large and one standard deviation is enough to engulf the other groups. The authors support for this claim comes from the line: "ANOVA, df = 2, 98, F = 7.91, p<0.001; Tukey's HSD post-hoc test". The F statistic is far too large to correspond with the data presented in Figure 2. The summary statistics in the Supporting Information also suggest that means and SD associated with the groups will not differ. I call upon the authors to clarify and post the data for transparency.
RE: Figure 2 does not support chloride claims
joshua_thienpont replied to Garbarino on 16 Dec 2013 at 14:26 GMT
The analysis of variance that was conducted in order to characterize the difference in chloride concentrations between the three a priori-defined groups were run on log-transformed data, as the original data were determined to be not normally distributed. The values for the ANOVA presented in the paper are correct based on these transformed, and subsequently normal, data. The Tukey’s HSD post-hoc test then determined that there was a significant difference between the drilling sump and control lakes (adjusted p = 0.0007) and between the drilling sump and thaw slump lakes (adjusted p=0.05). The data that are presented in figure 2, and summarized in the supporting information are the untransformed raw data collected during sampling. The raw data were presented in this format in order to provide continuity with other studies of the water chemistry in the region (such as Kokelj et al. 2005) that presented untransformed data.
In addition to the claims based on chloride concentrations, our conclusions on the differences in water chemistry between the lake groups were based on the entire suite of water chemistry variables which were sampled and are summarized in the principal component analysis and tested using an analysis of similarity. The fact that a similarity percentages analysis identified chloride as being an important variable for determining the differences between the significantly different groups based on ANOSIM suggested to us that chloride is one of a number of important variables that separate the water chemistry of these lakes.
The complete water chemistry dataset is currently being prepared for upload to the NWT Discovery Portal http://nwtdiscoveryportal... where it will be freely available to all interested parties.
Sincerely,
Joshua Thienpont, PhD on behalf of the authors
RE: RE: Figure 2 does not support chloride claims
Garbarino replied to joshua_thienpont on 17 Dec 2013 at 23:35 GMT
The authors have acknowledged that the data presented in Figure 2 (means and standard deviations) were not those on which the analyses were performed. Therefore, the authors have now accepted that Figure 2 is incorrect. The correct figure should remove the letters A/B or plot the log-transformed means and standard deviations. Only then does the figure match its caption and the paper text.
The comment from the authors also reveals that the analysis is on geometric means because the data are log-transformed. The data plotted in Figure 2 seem like arithmetic means, and so this again is a mismatch. Why use geometric means?
It is sad to hear that the data are not yet available. I want to see results presented from a nonparametric bootstrap analogue of the ANOVA that does not require any assumption of normality and is a test about the untransformed means shown in Figure 2.
Discussion about a weak result (R = 0.31) from a very complicated multivariate analysis with 19% contribution of Cl is irrelevant to my comment. You can also draw no conclusion from your PCA because many of your water chemistry variables are likely highly correlated, like Cl and K and Na. Correlated variables remain correlated after PCA, meaning you cannot resolve effects of different variables. All you can do is summarize them onto new axes if multicollinearity is a problem. Again, there is no mention of the extent of multicollinearity in the paper and there is good reason to suspect that this is high and confounding.
S Garbarino
RE: RE: RE: Figure 2 does not support chloride claims
joshua_thienpont replied to Garbarino on 28 Dec 2013 at 10:56 GMT
The caption that accompanies Figure 2 as well as the text of the manuscript explicitly state that the figure represents the mean and standard deviation of the data and that the determination of differences in the three a priori-defined groups are based on ANOVA on log-transformed data. So no, it is not incorrect. The use of analysis of variance following transformation to satisfy the assumption of normality combined with ANOSIM and SIMPER analyses and the appropriate ordination method are regularly used in a wide range of studies. This standard workflow is commonly used for the analysis of multivariate water chemistry datasets and was deemed to be the correct, appropriate approach by the authors, the three expert reviewers, and the editor of the manuscript. Future analyses of this and other datasets will necessarily take into account a variety of non-parametric techniques when the assumptions of more commonly-utilized analyses are not met.
Regardless of these discussions, a simple examination of the chloride concentrations in the sump-affected lakes compared to the slump lakes and the control lakes, fully confirms our conclusions. And of course, the limnological data are just a supplement to our detailed paleolimnological study – which of course was the main thrust of this entire paper.
Joshua Thienpont, PhD
On behalf of the authors