Detecting Associations between Archaeological Site Distributions and Landscape Features: A Monte Carlo Simulation Approach for the R Environment
Round 1
Reviewer 1 Report
The paper aim was to apply Monte Carlo methods to test for association between Lower and Middle Palaeolithic sites in Hampshire and Sussex, UK, and quarries recorded on historical maps. The authors present this approach as an alternative to traditional statistical methods such as X2 and regression analysis. Although the application of these methods is not novel in archaeological research, they are still poorly explored. Although their overall aim is clear, it perhaps misses the possibility to address a larger problem. In the start, they suggest they wanted to tackle a fundamental problem in archaeological research, i.e. the incompleteness of the archaeological record and the potential of regional data to understand larger patterns of human action. But later, the aim is to focus on a site location problem. Nonetheless, this article an interesting contribution to the development of advance statistical methods in archaeology, and its application to diverse datasets and regions.
Lines 58 and 59 reads “Besides the difficulty of locating fragments of lost geographies and archaeological remains there remains the major fundamental issue of relating artefacts to sequences.” This sentence is a bit confusing, I would say that “lost geographies” is an interpretative category, usually, we don’t find a “lost geography” but rather we (re)construct it, it is not a directly unit registered in the field. On the other hand, “archaeological remains” -while also classify and categorize – are empirical units, observable and registered in the field. I think the authors are mixing different levels of the analytical process here. Also, consider changing the word “remains” since it is used with different meanings too close.
The paragraph from lines 58 till 70 can be improved and therefore the ‘justification’ of the statistical methods used. The authors based their decision on the basis that traditional statistical methods are insufficient or ‘not intuitive’ enough for researchers inexperience with statistics. That is a point that has been rising since the early 2000s by many researchers, most of them from the UK, and therefore perhaps not strong enough to justify the use of MC methods. I would suggest the authors to also include a sentence or a paragraph on the potential and advantages of MC methods in archaeology based on previous research. In this sense, the paper will not come as a repetition without much citation of what has been already established, but rather a strong new contribution.
Another point is, while the authors use R for the analysis, it is not clear if the datasets used is also open access and available for other researchers to replicate their analysis. I think it will be very positive for your research to make your data also open access.
It does not seem to me that the sections on X2 and regression analysis are fully necessary for the argument of the paper. They add volume but not analytical argumentation, those sections can easily be summarized in half a page. Also, these sections do not reflect the extensive archaeological literature on these topics and the ‘pros and cons’ of its uses for different archaeological contexts. I would suggest reconsidering the validity and overarching importance of these sentences for the paper aims.
Lines 266 to 270 state “Sites ranged from single lithic findspots with no accompanying information, to important sites like Warsash where large lithic assemblages have been recovered and stratigraphic sequences have lately been reconstructed”. I wonder how are the authors dealing with the uncertainty of regional distribution based on such different sites for their overall archaeological interpretation of the regional patterns. Is the functionality of each site relevant for the study? This is, a place were one axe was found cannot be compared with “important sites like Warsash where large lithic assemblages have been recovered and stratigraphic sequences have lately been reconstructed”. Particularly if the underlying aim is to reconstruct ancient human patterns. I think it would be important to explicitly define what a site is and how different types of sites are being classified and considering within the analysis.
Just for extra clarification the line #300-301 “If we find no association…” could be complemented with what association you are referring, i.e. “between known locations of quarries that were worked during the period”.
Section “3.3.2. Adding buffers to account for uncertainty” is not very clear on its explanation, perhaps an image with the grids/circles could help the readers that are unfamiliar with the UK grid system to better understand what is it that you did.
Section “3.3.3. Random sites generation”, you do not explain how many random points were calculated. Are the random sites equal to the observed ones? How many sites per circle?
Lines 262-363 “The final part of the analysis involves overlaying first the real sites, with their buffered 362 probability zones…”. First, it is not clear if these probability zones are connected to the first buffer zones or with the random points/circles. Also, I suggest modifying the “real sites” for “observed/recorded sites”.
I think that when it comes to descriptive writing the paper is a bit confusing. I understand the value of being concrete and clear, but it seems that in these descriptions information is missing to present a full and clear picture of what you did.
Sections “3.4 MCSites: A Monte Carlo analysis tool for R”, “3.4.2. Data import and analysis preparation in R” and the rest up to “3.4.6” seem too technical, they are relevant for the methods chapter of a master thesis, however, for a paper consider moving it to the appendixes.
Lines 519-520: This is too informal and not very scientific, what do you mean by “a good reason to believe in the selected explanatory variables”?
Line 520: When you write “regression approaches are clearly superior to this simple test”, which simple test are you referring, X2? Because it sounds as if you are referring to MC.
In the conclusions, the authors state that “The Monte Carlo approach described offers a rapid and effective way of testing association between archaeological site distributions and landscape features of various kinds, like superficial geological deposits, land cover or vegetation types, or in this case, quarries.” (lines 514-516). However, in the preceding discussion, they did not explain any other environmental or paleoenvironmental variables beyond the ‘quarries’ (which it is not an environmental variable since in the line #279 they explicitly state that they refer to the mining places). In this sense, the statement that “The Monte Carlo approach described offers a rapid and effective way of testing association between archaeological site distributions and landscape features of various kinds” have no actual solid ground, at least not from the case presented by the authors. I would strongly suggest making this clear to 1) provide a stronger conclusion of the possibilities and dimensions of MC approaches in archaeology, and 2) avoid vacuum statements.
Line 535: I don’t think that “The approach described also offers a way to account for uncertainty and imprecision in site location.” You describe one of the aspects that from your understanding could produce an element of uncertainty for site location. Uncertainty in such a database could be present in many different variables. Also, your treatment of uncertainty is quite subjective and not quantitative. See for example Crema (2012). I would recommend building a stronger case, both in the discussion and conclusions, on how your use of the MC approach contributes to accounting for uncertainty.
Line 545-546: I don’t think this is necessarily true. It is relative to the community you are referring to. For example, among the CAA community, I think the use of R is quite common.
Line 548-551: This sounds like an oversimplification. However, not for a paper since it invalidates decades of research and application of R in archaeology, see for example textbooks such as Connolly and Lake (2006) or Carlson (2017).
Line 565: consider changing the repeated word “clearly”.
Line 572-573: “the approach we have demonstrated in this paper is 572 robust and simple to apply, and deserves to be better known”. This does not sound like a strong scientific reason, consider modifying it.
As a final reflection, I think this is an interesting paper and should be published. It is clear that the authors are aware of archaeological and general scientific research on this topic, and they properly apply the method to an interesting case study. I would mainly suggest three major points for improvement: 1) try to be coherent on the promises you make along with the paper and how they are represented in the discussion and conclusions (e.g. the use of environmental variables and paleoenvironments). 2) Avoid the colloquial language when stating your points, always look for a scientific style, and base your comments on your presented analysis and results (see various comments above). 3) Avoid unnecessary information or highly technical descriptions that can either be a reference to a thesis or add in the appendix.
Author Response
Please see attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
This is an excellent publication that I strongly recommend for publication. The writing is clear, the arguments well-focused, and the presentation of an open source method for implementing the Monte Carlo test is welcome and will be a tool that many archaeologists may choose to use.
I rank the quality of presentation less than "High" only because there are some edits to the instructions for running the code required.
I rank the "Scientific Soundness" less than "High" because I remain concerned about the caveats that the authors themselves present concerning the sample of sites and the bias in site identification. This is really a conundrum that I don't think can be resolved. Therefore, it is scientifically sound in that the article presents a reasonable test given the nature of the sample. However, the sample itself is problematic. It will remain up to other scholars implementing this method to present reasonable arguments for any inferences drawn from using this tool.
Specifically: I encountered a consistent problem implementing the sample data and code as instructed in the article.
Using R-Studio (v. 1.3.1056) and R 4.0.2 (downloaded from https://cran.r-project.org/) I received an error message that rangen.R could not be found. When I moved rangen.R from the "scripts" folder into the higher-level MCSites folder the problem was solved. I suspect that the call for rangen.R in the code needs to specify the "scripts" file. Correction of this issue should be made either in the code or in the instructions prior to publication.
More specific instructions concerning how to direct R-Studio to the working directory would be useful for scholars who have not used R-Studio previously.
There is discussion of a manual workflow using ArcGIS. More specifics for using that workflow would be welcome, perhaps, for those unable to implement the R workflow (though I wouldn't make this a requirement, as it is not the goal of the article).
Author Response
Please see attachment
Author Response File: Author Response.docx
