Dear Dr Leblans,

Many thanks for the revisions. You have managed to address most of the major concerns raised by referees. However, there is one remaining issue regarding the statistics and how ‘plots’ were used to apparently increase replication (i.e. pseudoreplication). I include below the comments of one of the referees who has seen your revised manuscripts and explains in good detail why he/she thinks that inclusion of ‘plots’ is problematic.

The comments are substantial enough to require changes in your statistical methods, or at the very least a clearer explanation of how tests were applied. I consider these to be major revisions, but hope that you will be able to address all of the concerns raised, and adjust the manuscript accordingly.

With best regards,
Jens-Arne Subke


Referee comments:
“I think the authors have done a good job addressing most of my comments, and they certainly made a lot of changes. There is only one comment that I am not happy with the response to: Comment 3. My concern was not that plots had been included as random rather than fixed variables, but rather that they were included at all. There were 18 plots, and there were multiple samplings per plot, and at multiple times. Obviously, samples taken within the same plot are not independent of each other - they are closer to each other than they are to samples in different treatments, and are likely to have similar values . The appropriate thing to do would be to use their mean values per plot, or, in the case of multiple time periods, run either a time series analysis or run separate analyses. But if plot was included as a variable, then they MUST have counted multiple samples per plot as independent, or they would have run out of degrees of freedom and not been able to address the question.

I think they ran something like this:


Y=seabird colony+ tephra depth plot + sample(plot)
or Y=seabird colony + tephra depth + plot + year
or some combination therefore.

These would not be valid, because the degrees of freedom would have to be >18 for that analysis to run (since plots alone chew up 17 degrees of freedom), and there are only 18 independent points (plots)

I would ask them for an explicit model, complete with degrees of freedom associated with the F-statistics (numerator and denominator). If it was done like this, there needs to be either a very good explanation of why samples within plots are truly independent (otherwise one could simply take more samples per plot and gain power!), or it needs to be rerun as I suggest in the top paragraph.. The fact that plot was retained in their models (after the main treatments had been accounted for) supports the notion that they samples within plots were very similar to each other!

On that note: I don't understand why Table 1 includes sums of squares but no F-values (without SS for the error term, these are impossible to interpret). Furthermore, they appear to have included the numerator degrees of freedom ((2-1)*(2-1)=1 for all of them), but what we really need to know is the denominator degrees of freedom - that will tell us something about the total number of experimental units.”

Dear Dr Leblans,

Many thanks for the last version of the manuscript. I agree with your response and the changes you have made. Even if the conclusions were not affected, it is good to have analysed the data with appropriate statistics.

You still refer to the generalized linear model, rather than ANOVA in the cations of Tables 1 and 2, please correct. Note also that Table 1 is truncated - I assume that all is fine with the detail not visible here, but please check carefully before submitting the final (non-truncated) version.

Kind regards,
Jens-Arne Subke