Heterogeneity in US Farms: A New Clustering by Production Potentials

Round 1

Reviewer 1 Report

The article takes a closer look at the characteristics of U.S. state agriculture from the perspective of its production potential. A useful article, especially for the non-U.S. reader. That being said, the variables listed in Table 1 should be vaguely explained. Their understanding for the non-U.S. reader requires additional study. This is especially true of the variables: AGL, AGLPO, FT, TC, UT.PO. This should not be the case. I also suggest, for example, to include information on what measure is 1 acre (in relation to 1 hectare, how many square meters it has).

Research correct article. It deserves to be published.

Author Response

Please see the attachment. 

Author Response File: Author Response.docx

Reviewer 2 Report

The topic of the paper is actual. I suggest to make attention to the word "classification" in the title and the cluster analysis in the paper. The investigation is structured well: I think that the preprocessing step can be separated. The authors perform agglomerative hierarchical cluster analysis to group the data by similarities but it is not clear about the used methods or used tools. The investigation is described sequentially. The literature covers a long period of observations.

Technical remarks: row 334 -> the word "the difference" is dublicated

                               table 3 -> double brackets

The topic of the paper is suitable for Agriculture journal. 

Author Response

Please see the attachment. 

Author Response File: Author Response.docx

Reviewer 3 Report

This paper characterizes farm diversity in the United States of America at the county level using hierarchical cluster analysis based on a potential poor of variables that is slightly different from existing studies, which have done the same thing. This is a worthwhile study, but to be published needs deep and extensive revisions. Before listing suggestions details, the lion's share of information on how to address the below comments can be found by reading the Methods sections of the countless studies from the "farm typology" literature. A good journal for this is Agriculture Systems and, as noted in the References here, in MDPI's Agriculture. It is recommended that the authors carefully consult this literature to inform and improve upon the terminology and methodology of this paper. Generally, this is a good practice when starting out in any corner of multivariate statistics. With significant and extensive revisions, this paper could publish in Agriculture. 

 

------

Suggestions for improving “Heterogeneity in US Farms: A New Classification by Production Potentials”

 

Abstract:

·      Very incomplete

o   No justification given

o   Why United States?

o   Why is this paper necessary, and what knowledge gap does it fill?

o   There is no research question apparent. What is the question and why is it important?

o   Is the ‘clustering’ the primary objective? Other studies have referred to this as a ‘typology’. Is this this same thing?

o   Is minimizing diversity within a cluster really the most important takeaway here? Why would readers be interested in a ‘cluster’ instead of some other conceptualization that the clustering could help develop?

o   The policy-relevance here is a little unspecified. Why focus on shocks instead of some other aspect of policy response? I'll expect "shocks" to be extensively discuss later if this is the main takeaway.

o   What kinds of “shocks” are you referring to?

§  Price shocks? Climatic shocks? Technological shocks? Etc…

o   The first sentence of the Introduction is very relevant for the Abstract and should probably be in there in some form.

Introduction

·      The logical flow from the first to the third paragraphs is not entirely clear. Please make the rationale for the paragraphs more clear. For example, the first talks about the existing patterns in US farming that have been identified, but then, how does this lead to the next paragraph? It is not clear to me what the train of logic is here. Also, the reasons for the third paragraph are not entirely clear to me. Why is this information important? Again, if there was a stated question(s) or hypothesis that derives from a research gap, the logic would be more clear. While the fourth paragraph is more clear, again, shouldn’t the focus be on farm ‘type’ or ‘typology’ or ‘classification’ rather than ‘clustering’? Clustering is a statistical method, not a policy product or target of policy.

·      Paragraph 5. Why take this approach (i.e., w/o considering state lines or resource regions)? Please justify why this was done.

·      Paragraph 6. Here you are focusing a lot about economic aspects of farms and the effectiveness of agricultural and economic policies as a primary reason for building the classification. However, what happened to the “shocks” referred to in the abstract as the primary takeaway from the research? I hope all of this is revisited in the Discussion section, no?

·      Paragraph 7 seems out of place and a little irrelevant.

·      Paragraph 8 seems better aligned with Paragraph 2. Please streamline.

·      Paragraph 9: it is not clear how this paper will “redefine the concept of delineating agricultural subregions”. As far as I can tell, this paper is using cluster analysis to characterize agricultural diversity in the United States of America. This is a widely used methodologically. Conceptually, and in terms of spatial organization, this does not seem like a redefinition of anything. Many, many classification and typology studies have been performed using this technique and methodology. Please clarify or state differently (hedge).

Materials and methods

·      Yes, but why does THIS paper use hierarchical clustering? Many similar studies also use K-means clustering. Why not this type? Please justify beyond stating that other studies use it.

·      It is not clear how the approach of Sommer and Hines’ had a “disadvantage” simply because one type of variable “influenced” more heavily? Perhaps, this simply reflected the lack of statistical clustering in the farm types and, therefore, was actually MORE accurate of the actual farm diversity (or lack of) and therefore, more advantageous. When you talk about your study, by contrast, being more “accurate”, do you mean you selected fewer variable classes so that you could have better statistical clustering "accuracy"? Or, that your clusters more “accurately” characterized existing farms and farming regions? Please be clearer and more careful in the explanations here about what is more “accurate” and therefore, optimal given your research questions and objectives. Again, carefully reading the literature of "farm type" studies will help with this language and terminology.

·      Paragraph 2 of the Materials and Methods needs to be revised greatly.

o   Some of the wording is imprecise and some is incorrect on statistical grounds.

o   Another important point is that this study doesn’t really identify heterogeneity within agricultural systems; rather, it identifies similarities in characteristics. Clustering in this context is a means of simplifying farm complexity into a limited number of similar types. It does not expound on farm complexity as such. In other word, it is a type of dimensionality reduction tool (like PCA, for example). It is agglomerative. 

o   There is no rationale for WHY the list of variables was shortened to 15. Did you use statistical methods or a lasso procedure or other statistical or theoretical reasons for variable selection? Or, were variables chosen only so they would cluster well (e.g., introducing all sorts of theoretical and methodological problems)? The rationale for variable selection is unclear, but this is an important part of the study study design and rigor.

·      Table 2; Figure 1.

o   I would include the Figure in the text, but put the table as supplementary material. As is currently present, it is repetitive.

·      In paragraph 5, a small point. I’m assuming the clustering algorithm did the agglomeration, rather than “we”. Please clarify. Again, reference other studies.

·      Paragraph 6. I do not think there needs to be so extensive an explanation of the use of Wards method. Just say, “We used Ward’s method, a robust…to develop the cluster dendrogram”.

·      There is far too much description of the clustering process.

·      Figure 2. There is too much here. Please simplify. On the first part, you do not need to black boxes at the bottom of the dendrogram, which you explain as showing nothing, really. Again, reference other studies--there are some good ones in your bibliography. Also, please delete the first dendrogram and keep the second one, which has the colors. And include the cutoff point for the clusters. However, the cluster colors are too similar. Please choose another color scheme to better distinguish clusters. If you use the dendrogram (my personal preference for hierarchical clustering) you do not need the radial plot. Another possible illustration would be a constellation plot, but this too would be repetitive. Please just pick one (e.g., modified dendrogram).

·      These figures are probably better positioned in the results section instead of the materials and methods section.

·      There needs to be a more rigorous method of deciding the optimal number of clusters. There are many (e.g., using the elbow method of different criterion plots; the Silhouette Method; the Cubic Clustering Criteria Method; and many more. The number of “optimal” clusters must be decided based on clear and justifiable criteria. (see the many farm typology studies available).

·      In the Methods section, the paper should explain how the clusters are then validated (through the description started with in the results, etc…).

Results

·       Again, why THESE particular attributes (variables)? There needs to be a justification for this more rigorous than just they provided the best statistical fit. See literature.

·       Figure 3. Please show either the dot map or the heat map. Both together is repetitive unless the background (base) map is greatly simplied, in which dots and heat map can be superimposed. But again, simply please.

·       It isn’t the “largest cluster”, rather, this cluster contains the largest number/percent of counties—if that is the unit of analysis.

·       Figure 4. Again, please use either dots or heat maps, not both.

·       Figure 5. The turquoise color does not show well on the map. Please consider a better color for both dots and background Also, consider just using black and white maps and color dots for the clusters (again, with more distinct colors). Maybe even use a national-level county map, since that is the unit.

·       Figure 6. Please use either or, and do NOT alter the map dimensions compared to other maps. The images in Figure 6 are too skewed compared to the other maps.

·       Consider doing some other type of statistical analysis (e.g., analysis of variance) to show statistical differences in traits among the clusters. This would provide greater insight into the results and findings and would pave the way for a more thorough Discussion section. See literature for guidance.

Discussion

·      This section is underdeveloped. Part of this goes back to the lack of an original research question.

·      In the Discussion, please revisit the major objectives of the study and address the main questions and hypotheses of the study.

·      For example, how does this study contribute to existing literature?

·      How does it address a research gap?

·      Was there anything unexpected or particularly interesting in the results that you found?

·      What are the major limitations of this study?

·      What are the impending or unresolved issues that warrant further investigation?

·      What are the major policy implications (as referred to in the Abstract and Introduction sections) of the study?

Conclusions

·      There is no Conclusions section. This seldom happens, but when it does, it usually occurs only when there is an exhaustive and detailed Discussion section. This is not present either.

·      So, revisiting the title, what is the takeaway of this heterogeneity? What is the takeaway of the production potentials? How will this understanding lead to broader impacts in the political-economic literature and/or applied world of agricultural economics? There is very little in this paper that refers to economics—nothing about crops or prices or volatility or revenue/farm income, etc…this reads more like a farm characteristics classification or typology study, but there is very little discussion in the realms of policy actions or other applied ends.

 

In sum, the bones of a publishable paper are here. But the authors should be better informed by the extensive literature to improve the methods. They should also GREATLY expand on the research contextualization, discussion of findings and addressing of questions, testing of hypotheses, and with a more developed conclusions sections.

Author Response

Please see the attachment. 

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The authors should be congratulated for a fine set of revisions. The paper is now substantially improved. The 'cut-off' decisions about how many clusters to include from the dendrogram is still based on very subjective criteria (e.g., the sweet spot?) and therefore may not provide as useful or accurate a representation of farm complexity as more supervised techniques. Also, k-means (or medians) clustering can be conducted using unsupervised (exploratory) means, depending on the software, through iterative runs of the algorithm. In addition, we continue to hold that the term "clustering" provides a poor conceptual focus as it causes many to think of spatial clustering, which is not a focus of the paper. Clustering is a statistical method, not a farming conceptualization. Types or classes of farm characteristics are still better (and more accurate) representations of what the study is trying to do, but will let it go. The justification that the "clustering" term will generate more citations and is therefore more legitimate raises some questions about scholarly rigor and integrity. But in this case, these are mostly minor considerations. Overall, this is a significant improvement and the authors should be congratulated.