|The authors have addressed most of mine and the other reviewers’ comments, but it seems that my suggestion to conduct a more in-depth statistical analysis has been over-interpreted. The authors have now added a large amount of statistical analyses, mixing parametric with non-parametric statistical approaches. It is indeed true that there is a debate on which one is the appropriate approach to analyse ordinal data. However, I believe that in the presence of such debate, the authors should take a stance and choose the approach they deem more appropriate for their data and present their arguments for such choice. My opinion in this regard is that this approach should be non-parametric, as ordinal data can often have error structures that are not normally distributed, and thus violate common assumptions of many parametric approaches. The authors’ themselves in line 471 state “Since we have an ordered variable, and the distances between the categories are not the same […]”. In line 436 the authors state that “Kruskal-Wallis rank sum test is more powerful because it uses the mean of the rank to assess if there are differences in the responses of different groups (Agresti 2002, Magnifiaco, 2016), not requesting further assumption about the distribution of the data, although the test is fit for small samples in which there are not normal distributions”. If the authors think that this test is better than the one previously presented, why then not just using this one? It would slim down the analysis (e.g. the analysis would then consist of Chi-sq tests + Kruskal-Wallis test + CA*). Anything between lines 482-503 and related results can then be removed (together with Tables A4, A5, A8). |
Moreover, the basic concepts and principles of statistical analysis (such as those behind hypothesis testing) do not need to be presented in a research paper, as the reader is expected to understand the basics behind statistical testing (i.e lines 426-435 and 438-440 can be removed). The statistical methods section should serve to name the tests used as well as any additional information useful for the reproducibility of this study. It is indeed not always necessary to explain in detail what a test does, unless the author argues for one test over the other (and thus needs to argument the “why” behind the choice).
*I had previously recommended reconsidering the use of CA because there are, in my opinion, other more straightforward methods to show these difference (even by simply plotting the data). However, the choice in this sense falls on the authors.
- Unfortunately I forgot to mention this at the previous round, and I apologise for bringing this up only now, but naming the variables Q1, Q2, etc. makes it extremely hard to read the tables presented in the paper (as the reader has to always go back to the survey form). I recommend assigning a more descriptive name to the variables.
- Lines 395-398 could be removed as it’s not relevant to the end of reproducibility of your results. It would be enough to write that the statistical analyses presented in this paper have been conducted using the software for statistical computing R.
- Line 399: I would use “ongoing” instead of “never-ending”, as it seems too strong.
- Lines 410-411: I think this sentence is superfluous.
- Line 414: While I completely agree that plotting Likert-type scale like the authors did is one of the best ways to plot these data, I wouldn’t consider it an analysis per se, but rather a data exploration (fundamental step before starting the actual data analysis).
- Lines 418-420: In light of what the authors write here, with which I agree, I would remove this sentence and the corresponding Tables A2 and A3. In my opinion, it doesn’t add anything relevant to the analysis.
- Line 505: “selected questions”, do the authors’ mean the main questions relevant to the study?
- Line 530: what are the age segments corresponding to “young”, “mature” and “old”?
- Table 4 could be moved to supplementary
- Lines 512-514 sentence unclear (missing verb?)
- Line 558: Figure 3 does not show drought
- Line 562: “There are ten years since I had serious problems every year” maybe should be “In the past ten years, I had serious problems every year”. Maybe “achieved” should be “purchased”? Just checking because I am not aware what was meant in the original language
- Line 578: “Stakeholders’ ” and not “stakeholders’s”
- I noticed the authors changed school head into school director throughout the papers, but in the figures it still says school head. I would be consistent with school director throughout.
- Figure 8 caption: I would just keep “Stakeholders’ past experience with natural hazards”, as yes-no questions are not on a Likert scale
- Line 773: there seem to be some issue with the numbering of figures. E.g. in page 35 it should be Fig. 10, but it’s labelled as Fig. 2. Similar in page 36.
- Figure 9: I assume the blue and red colours for “Q6 damage” in Fig. 9 represents no and yes, respectively, but it should be stated in the caption.
- Table A1: I think it would be good for the sake of consistency to assign a-h values for all the items, not only for the those in the first questions
- Table A7: it’s common practice to not write anything if a value it’s not significant (i.e. no “ns” needed)
- Table A9: the authors don’t mention the use of logistic regressions in the methods section, but their results are presented here. Results of a logistic regression are generally reported either as beta coefficient with related significance (p-value) or as odd ratio with related confidence interval. It is also not clear between which variables the regressions were run (but it seems between all of them) and the significance should be corrected for in case of multiple testing on the same variable.
- I would recommend an additional proofreading of the paper.