Global Hemp Production (2013–2023): The Roles of Harvested Area and Fertilizer Use
Débora Paulino
Abstract
Industrial hemp is frequently discussed as a potential crop for agricultural diversification and bio-based production systems, yet cross-country evidence on its production dynamics remains limited. This study examines the relationship between hemp production, harvested area and national fertilizer consumption using an unbalanced country-year panel covering 2013–2023. Pooled ordinary least squares and panel-data estimators are used to investigate associations between production scale and aggregate agricultural input use, with complementary specifications examining yield. The results indicate a strong positive association between harvested area and hemp production. National fertilizer consumption is also positively associated with production, although its estimated relationship differs across model specifications. These findings suggest that changes in cultivated area were an important correlate of hemp production dynamics during the study period. However, because fertilizer consumption is measured at the national rather than crop-specific level, the results do not directly establish fertilizer requirements or environmental performance in hemp cultivation. The study contributes cross-country evidence on the structure of hemp production while highlighting the data limitations that constrain assessments of crop-specific input use and productivity.
Review reports
Reviewer3 (AI review)
AI Review (reviewer3.com)
Comments (12)
1. Unreported yield regression, random effects, and Hausman test results — Critical
A complementary yield-based specification serves as a robustness check and focuses on productivity rather than production scale. In this case, the dependent variable is the natural logarithm of hemp yield, defined as output per hectare:
Multiple sections state that panel regressions for hemp yield, random effects models, and Hausman tests were estimated and used to draw conclusions. Specifically, Section 2.3 (page 12) defines the panel regression equation ln(Yield_it) = α_i + δ_t + β₂ ln(Fertilizer_it) + ε_it, and Section 4 (pages 18–19) discusses findings from it ("Once cultivated area and unobserved country heterogeneity are controlled for, fertilizer consumption is positively but modestly associated with hemp production and productivity" and "reinforced by the weaker and less robust relationship observed between fertilizer use and yield"). Likewise, Section 2.4 (page 13) states that "random effects models are estimated" and "Model selection between the random effects and fixed effects specifications is guided by the Hausman test." However, Section 3 (Results) only reports Pooled OLS and Fixed Effects models for hemp production (Table 2) and never presents or mentions any regression results for yield, random effects estimates, or Hausman test statistics.
2. Contradiction on whether hempseed data are analysed as complementary evidence — Minor
availability is more consistent across countries and years. Hempseed data are retained in the dataset and analysed separately as complementary evidence, but are characterised by a smaller number of producing countries and more frequent missing observations, particularly for fertilizer consumption.
Section 2.1 (page 3) states that "Hempseed data are retained in the dataset and analysed separately as complementary evidence," and Section 3.2 (page 16) states that "hempseed results are presented as descriptive complementary evidence." However, Section 3.2 (pages 15–16) simultaneously states that "all Hempseed observations are excluded from the fertilizer-based panel dataset" and provides no descriptive statistics, tables, figures, or quantitative numbers for hempseed anywhere in the manuscript.
3. Inconsistent units for hemp yield across text, table, and figure — Editorial
value. The production elements include production (measured in tonnes), harvested area (measured in hectares), and yield (measured in tonnes per hectare). Country names are
In Section 2.1 (page 3), the text states that yield is "measured in tonnes per hectare" and "calculated as the ratio of production to harvested area" (where production is in tonnes and area is in hectares). This unit matches Figure 3 (page 10), where "Log yield" spans roughly from -2.5 to 2.5 (corresponding in natural logarithms to approximately 0.08 to 12.2 tonnes/ha). However, Table 1 (page 4) reports "Yield (kg/ha)" with a mean of 2,825, a minimum of 71, and a maximum of 12,600.
4. Invalid comparison of elasticities to infer extensive vs. intensive growth — Critical
level. When unobserved country-specific effects are controlled for in the fixed effects model, the estimated elasticity increases to 0.545 and remains statistically significant (p < 0.01). This suggests that, within countries over time, increases in fertilizer use are associated with higher hemp output, but the magnitude of this effect remains substantially smaller than that of harvested area.
The central conclusion that global hemp expansion during 2013–2023 was primarily driven by scale (land expansion) rather than input intensification is inferred by comparing the magnitude of the area elasticity (β_1 = 1.139) to the fertilizer elasticity (β_2 = 0.545) in Table 2. However, comparing partial elasticities (which measure the percentage response of output to a 1% change in each variable) does not establish which factor actually drove historical output growth over the 2013–2023 period, because historical growth contribution depends on both the elasticity and the actual historical change in each input (i.e., Δln(Production) ≈ β_1 Δln(Area) + β_2 Δln(Fertilizer)). Moreover, in any standard agricultural production function, an elasticity of output with respect to land near 1 simply means that doubling land at constant input intensity approximately doubles output; simultaneously, a fertilizer elasticity of 0.545 is very large in an agronomic sense (implying a 10% increase in fertilizer is associated with a 5.45% increase in output holding hemp area constant, which is pure yield intensification). Even if the national fertilizer coefficient is attenuated by measurement error as the author notes on page 4, the true hemp-specific fertilizer elasticity would be even larger than 0.545, directly contradicting the claim that fertilizer plays only a minor role. To determine whether expansion was driven by area vs. input intensification, the author should perform a formal growth decomposition over 2013–2023 and revise the interpretation of the 0.545 elasticity.
5. Unspecified international statistical data sources and sample composition — Critical
The empirical analysis is based on a global country–year dataset covering the period 2013–2023. Data on hemp production are obtained from international agricultural statistics 2 sources and include two distinct hemp items: True hemp, raw or retted and Hempseed. These
The manuscript states that data on hemp production and national fertilizer consumption are obtained from 'international agricultural statistics sources' (such as FAOSTAT or World Bank data), but never actually names the specific databases, datasets, item codes, or retrieval dates anywhere in the text. Furthermore, for the regression sample of N=126 country-year observations, the manuscript does not report how many unique countries are included or list the countries and years covered. Because the reliability of country-clustered standard errors depends critically on the number of cross-sectional clusters (countries) rather than the total number of country-year observations, omitting the number of clusters prevents readers from assessing the validity of the statistical inference. The author should explicitly name the exact databases and series codes used, list the countries in the estimation sample, and report the number of country clusters in Table 2.
6. Omitted econometric results for Random Effects and Hausman test — Critical
cross-section and time series and ignoring unobserved country-specific effects. Second, random effects models are estimated to allow for unobserved heterogeneity across countries under the assumption that these effects are uncorrelated with the explanatory variables. Third, fixed effects models are estimated, allowing for arbitrary correlation between unobserved country-specific effects and the regressors. Model selection between the random effects and fixed effects specifications is guided by the Hausman test. When the null hypothesis of no
In Section 2.4, the author outlines a three-step estimation strategy involving Pooled OLS, Random Effects (RE), and Fixed Effects (FE), explicitly stating that model selection between RE and FE is guided by the Hausman test. However, Section 3 (Results) and Table 2 omit the Random Effects estimates and fail to report the Hausman test statistic, degrees of freedom, or p-value. To substantiate the methodological design and model selection described in the text, the author should include the Random Effects specification in Table 2 and report the exact Hausman test results in Section 3.3.
7. Undefined units and nature of the fertilizer variable — Critical
Information on fertilizer consumption is merged at the country–year level and is used as a proxy for agricultural input intensity. Fertilizer consumption is treated as a continuous variable and reflects aggregate fertilizer use within each country and year. While fertilizer
The manuscript uses national fertilizer consumption as its primary proxy for agricultural input intensity, yet it never specifies the units of measurement or whether this variable represents an aggregate national volume (e.g., thousands of tonnes of nutrients used across the entire country) or an agricultural intensity rate (e.g., kilograms per hectare of arable land). In Table 1, fertilizer consumption has a mean of 1,259, a minimum of 9.81, and a maximum of 27,444 without units. If the variable is total national tonnage rather than a per-hectare rate, it is a measure of overall national agricultural scale rather than input intensity; moreover, in the baseline log-log production regression (which controls for hemp area but not total national agricultural area), within-country changes in aggregate national fertilizer tonnage could simply reflect economy-wide agricultural land expansion rather than fertilization intensification. The author must explicitly define the fertilizer variable and its units in Section 2.1 and Table 1, and ensure that a per-hectare intensity measure (or a control for total national agricultural area) is used when testing input intensification.
8. Unverified visual inference of log production trajectories in Figure 1 — Minor
production paths. The Czech Republic and Ukraine show downward trends or increased volatility in later years, pointing to contraction or instability within their production systems. Russia, in contrast, exhibits relatively modest fluctuations around a stable but lower production level, suggesting continuity without significant expansion. These contrasting
In Section 2.2 (pages 6–7), the author categorizes countries into distinct production regimes based on visual inspection of Figure 1. However, each country facet in Figure 1 uses an independent, unstandardized vertical axis scale. For instance, Russia's vertical axis spans only 0.20 log units (7.10 to 7.30) and Ukraine's spans 0.075 log units (6.550 to 6.625), making minor fluctuations appear visually dramatic, whereas France spans 5.0 log units (7 to 12) and Spain spans 4.0 log units (3 to 7). Because the y-axes are not standardized and no quantitative metrics of volatility or trend growth are provided, visual comparisons of stability and volatility across facets (such as characterizing Ukraine as exhibiting 'increased volatility' when its total log range is less than 0.08) are misleading. The author should either plot Figure 1 with a uniform vertical scale (or consistent axis range span) or provide country-level summary statistics (mean, standard deviation, and growth rate) to support the regime classifications.
9. Omission of the specified fixed-effects yield regression results — Critical
Overall, the results indicate that once cultivated area and unobserved country heterogeneity are controlled for, fertilizer consumption is positively but modestly associated with hemp production and productivity, confirming that hemp expansion during 2013–2023 was primarily driven by scale rather than input intensification.
The headline claim and discussion assert that once cultivated area and unobserved country heterogeneity are controlled for, fertilizer consumption is modestly associated with both hemp production and productivity (yield), and that this interpretation is reinforced by a weaker relationship in the yield model. In Section 2.3, the author explicitly specifies a two-way fixed effects panel model for yield: ln(Yield_it) = α_i + δ_t + β_2 ln(Fertilizer_it) + ε_it. However, no estimation results (coefficients, standard errors, p-values, or R-squared) for this yield regression are ever reported in Section 3 or in any table; the only evidence presented for yield is a bivariate pooled scatter plot (Figure 3). Furthermore, because Yield = Production / Area, log(Yield) = log(Production) - log(Area); in the Table 2 Fixed Effects model where the coefficient on log(Area) is 1.139 (very close to 1), controlling for log(Area) in a log(Production) regression is algebraically nearly identical to having log(Yield) as the dependent variable, which yielded a substantial and statistically significant fertilizer elasticity of 0.545 (p < 0.01). The author must report the full regression table for the specified yield models (Pooled OLS, RE, and FE) to substantiate the claims regarding within-country productivity responses.
10. Inconsistent yield units between text and Table 1 — Minor
value. The production elements include production (measured in tonnes), harvested area (measured in hectares), and yield (measured in tonnes per hectare). Country names are
In Section 2.1 (page 3), the text states that yield is measured in tonnes per hectare, whereas Table 1 (page 4) reports yield in kilograms per hectare (kg/ha, with a mean of 2,825). Additionally, in Figure 3 (page 10), log yield ranges from approximately -2.5 to +2.5 (corresponding to 0.08 to 12.2 tonnes/ha), indicating that tonnes per hectare were used for the logarithmic transformation. The author should clarify and standardize the description of yield units across the text, tables, and figures.
11. Absence of statistical metrics for the bivariate yield relationship — Major
differences in agronomic conditions, management practices, and production systems. These patterns are illustrated in Figure 3, which depicts the relationship between fertilizer consumption and hemp yield for True hemp, raw or retted using pooled country–year data expressed in logarithmic form. The fitted linear trend suggests a positive association between fertilizer use and yield, indicating that higher fertilizer intensity is generally associated with higher productivity per hectare.
Claim 3 states that fertilizer consumption is positively associated with hemp yield per hectare, but that the relationship is weak and heterogeneous, relying visually on Figure 3. However, even for the pooled bivariate relationship in Figure 3, the manuscript provides no numerical slope coefficient, standard error, p-value, R-squared, or sample size (N). Visual inspection of a scatter plot without accompanying statistical metrics is insufficient to establish the strength or statistical significance of the association. The author should report the sample size, correlation coefficient, slope estimate, and p-value for the regression line plotted in Figure 3 (in addition to reporting the full panel regression results for the yield model specified in Section 2.3).
12. Complete omission of descriptive statistics for Hempseed — Major
and is characterised by substantial variation in scale. While production levels are positive in most reported observations, the availability of consistent fertilizer consumption data for hempseed-producing country–years is extremely limited. After applying the data-cleaning and consistency requirements necessary for the log–log econometric specification, namely, 15 non-missing and strictly positive values for production, harvested area, and fertilizer consumption, all Hempseed observations are excluded from the fertilizer-based panel dataset.
In Section 3.2, the author makes several empirical claims regarding Hempseed production—specifically that it is reported for a smaller subset of countries, exhibits greater discontinuity over time, is concentrated in a small number of producers, has positive production levels in most reported observations, and retains zero valid observations once merged with fertilizer data. However, no descriptive statistics (such as sample sizes N for production, harvested area, and yield, means, standard deviations, or number of reporting countries) are provided anywhere in the manuscript to support these assertions. Additionally, since national fertilizer consumption is merged at the country-year level using ISO3 codes, it is unclear why countries reporting Hempseed would have zero overlap with national fertilizer data when 140 country-years of fertilizer data were available for True hemp. The author should provide a descriptive statistics table for Hempseed analogous to Table 1 and explain why the country-year merge with national fertilizer data resulted in zero observations.
xPeerd (AI review)
AI Review (xPeerd — DAReview)
Summary
This manuscript investigates the relationship between hemp production, harvested area, and national fertilizer consumption using a global unbalanced country–year panel for 2013–2023. The core research question is explicitly framed as: “To what extent is fertilizer consumption associated with hemp production and productivity across countries over time, after accounting for differences in cultivated area and country-specific characteristics?” The study’s main strength is that it clearly distinguishes scale effects from input-intensity effects through a panel framework with country and year effects, and it is appropriately cautious in noting that fertilizer is measured only at the aggregate country-year level rather than for hemp specifically . The results consistently indicate that harvested area is the dominant correlate of hemp output, while fertilizer use shows a positive but smaller association with production and a weaker, more heterogeneous association with yield . The manuscript is coherent, methodologically conventional, and generally well written. Its main weakness is that several parts of the discussion move close to interpretive overreach relative to what the data can support, especially when linking estimated within-country associations to regulatory transition, sustainability, and production strategy. The empirical contribution is modest but valid as descriptive panel evidence.
Potential Major Revisions
- Strengthen the alignment between the stated contribution and the actual evidentiary scope.
The manuscript’s abstract appropriately cautions that “the results do not directly establish fertilizer requirements or environmental performance in hemp cultivation” , and the methods section likewise states that “fertilizer consumption is measured at the aggregate country–year level and is not crop-specific” . However, the discussion later advances broader interpretive claims, for example that “producers primarily adjusted hemp output by reallocating land rather than by intensifying input use within existing production systems” and that “long-term investments in input-intensive production systems may have been constrained by regulatory uncertainty and uneven market development” . These statements are plausible, but the model does not include direct measures of regulatory uncertainty, processing capacity, producer strategy, or investment behavior. As written, the discussion risks attributing mechanisms that are not identified by the design. I think the paper would be stronger if these claims were clearly reframed as hypotheses or contextual interpretations rather than implications of the estimates themselves.
- Address omitted-variable concerns more directly in the main results interpretation.
The model includes area and fertilizer as the main regressors, and the paper acknowledges that “other determinants of production, including labour, capital intensity, irrigation, varietal selection, pest management, and processing infrastructure, are not explicitly incorporated” . This limitation is important because the fixed-effects coefficient on fertilizer could still absorb time-varying structural changes correlated with both national fertilizer use and hemp output. The manuscript notes that “the empirical framework identifies within-country associations over time and does not establish causal relationships” , which is good, but the results and discussion should more consistently reflect that limitation. In particular, the sentence “increases in fertilizer use within countries over time are associated with higher output” is acceptable as descriptive econometric language, but subsequent interpretation should avoid implying that fertilizer itself is the operative driver absent crop-specific input data and additional controls .
- Clarify sample composition and possible selection effects from zero/missing-value exclusion.
The methods specify that “observations with zero or missing values for variables entering logarithmic transformations are excluded from the corresponding estimations” . This is standard for log-log models, but here it may materially shape inference because the descriptive section emphasizes “intermittent participation,” “entry–exit dynamics,” and strong panel imbalance . Excluding zeros likely removes a meaningful part of the hemp sector’s dynamics, especially for marginal producers. This matters because the substantive narrative emphasizes sectoral transition and instability. I think the paper should report more transparently how many country-year observations and countries remain in each regression, how many are lost due to log transformations, and whether the retained sample is disproportionately composed of relatively stable producers. Without that, the external interpretation of “global” hemp dynamics may be somewhat overstated.
- Improve transparency around the fixed-effects model selection and reporting.
The estimation strategy states that “model selection between the random effects and fixed effects specifications is guided by the Hausman test” and that the fixed-effects model is preferred when the null is rejected . Yet the excerpted text does not report the Hausman statistics themselves. Since the manuscript explicitly relies on the fixed-effects model as primary for inference, the paper should present the relevant model-selection outputs in the results, appendix, or supplementary materials. Similarly, because the discussion gives substantive weight to the coefficient differences between pooled OLS and FE, readers would benefit from fuller reporting of the full regression table, standard errors, number of groups, time periods, and any within/between variation diagnostics.
- Reconsider the interpretation of the elasticity above one for harvested area.
The manuscript is commendably cautious in stating that the area elasticity exceeding one “should not be interpreted as evidence of increasing returns to land in a technical production-function sense” . Still, the subsequent explanation invokes “shifts toward more suitable land, improved varieties, or better-aligned management practices” , none of which are measured. That paragraph is reasonable as conjecture, but it needs tighter wording because it may otherwise seem like a post hoc explanation for a superunitary coefficient that could also reflect compositional change, measurement error, omitted factors, or changing reporting patterns. A more explicit caution here would improve methodological credibility.
- The hempseed component currently contributes too little to justify its framing unless reframed more modestly.
The methods initially emphasize two hemp items, “True hemp, raw or retted and Hempseed,” and state that they are “treated separately throughout the analysis” . In practice, however, the econometric analysis is almost entirely about true hemp, while hempseed is excluded because “all Hempseed observations are excluded from the fertilizer-based panel dataset” . This is not a flaw in itself, but the framing should more clearly announce from the outset that the main inferential contribution concerns fibre-oriented hemp only. Otherwise, readers may expect a broader comparative analysis than the paper actually delivers.
- The manuscript should make a stronger anti-overclaim distinction between “sustainability compatibility” and actual environmental assessment.
The paper states in the discussion that the results “support the characterization of hemp as a crop capable of expanding without heavy reliance on increased chemical input intensity” and that this “reinforces arguments positioning hemp as compatible with low-input and environmentally efficient agricultural systems” . Yet the abstract already admits that the study does not directly establish “environmental performance” . I think these sustainability claims need to be softened substantially. Aggregate national fertilizer consumption is too indirect to support an environmental-efficiency implication, especially when no emissions, nutrient balances, land-use tradeoffs, or hemp-specific agronomic input data are modeled.
- Reproducibility is acceptable but still incomplete.
The manuscript reports the software ecosystem—“plm, lmtest, and sandwich packages in R” —and states that the compiled dataset is available “upon reasonable request” . For a paper relying on public data and relatively standard transformations, this is adequate but not ideal. A stronger reproducibility standard would be to provide the data assembly code, country harmonization rules, exact variable definitions, and replication scripts in a repository. Because the paper hinges on data cleaning and exclusions, reproducibility would materially improve confidence in the estimates.
Potential Minor Revisions
- Typographic and formatting inconsistencies
The manuscript contains visible equation-rendering artifacts in the econometric framework. For example, the sentence “The term αi\alpha_iαi captures unobserved country-specific effects that are constant over time, while δt\delta_tδt represents year fixed effects” contains corrupted inline notation that should be cleaned before submission . A similar issue appears in “β1\beta_1β1 measures” and “β2\beta_2β2 captures” . These are small but important presentation problems in a methods-heavy paper.
- Capitalization consistency
There is at least one instance where “hempseed” appears in lowercase at the start of a sentence: “hempseed production appears more episodic and country-specific” . This should be capitalized for consistency with the paper’s usual item naming conventions.
- Table presentation could be improved
Table 1 is informative, but the variable row “Fertilizer consumption 1,259 4,855 9.81 27,444 140” appears to omit a unit label or formatting separator compared with the other entries . This may simply be extraction noise, but if the table in the manuscript looks similar, the unit for fertilizer consumption should be stated explicitly and consistently. More broadly, all tables should define units, logs vs. levels, and sample sizes unambiguously in notes.
- Results reporting should be more granular
The results section reports key coefficients in prose, including area elasticity of 1.075 in pooled OLS and 1.139 in FE, and fertilizer elasticity of 0.190 and 0.545, respectively . This is helpful, but readers would benefit from confidence intervals in addition to p-values, especially because the paper’s substantive interpretation relies on relative magnitudes across estimators. If space allows, standardized reporting of coefficient uncertainty would improve readability.
- Clarify the phrase “conservative associations”
The methods state that measurement error is “likely to attenuate the estimated coefficient towards zero” and therefore “the estimated fertilizer elasticities should be interpreted as conservative associations” . This is directionally reasonable under classical measurement error assumptions, but the wording may be too confident because the merged aggregate fertilizer proxy could also capture other country-level processes. I would suggest using more careful phrasing, such as “potentially attenuated and difficult to interpret precisely.”
- Section-level redundancy
Several ideas are repeated across Results, Discussion, and Conclusion, particularly that production growth was “primarily driven by scale rather than input intensification” . The repetition is not severe, but some trimming would make the manuscript more concise and reduce the sense of interpretive overinsistence.
- AI content analysis
Estimated percentage of AI-generated content: low to moderate, approximately 15–30%.
The manuscript is highly regular in sentence structure, lexically polished, and stylistically uniform throughout, which can be consistent either with careful human editing or partial AI-assisted drafting. Several passages show formulaic transitional phrasing such as “These findings suggest…,” “This interpretation is reinforced by…,” and “These observations reinforce the view…” repeated across sections . The prose is also unusually even in cadence and hedging density, especially in the discussion and limitations sections . I do not see obvious hallucinated references or incoherent terminology, and the manuscript remains internally consistent. Therefore, if AI assistance was used, it appears mainly stylistic rather than substantive.
Highlighted potentially AI-assisted sections:
- Discussion paragraphs synthesizing regulation, land allocation, and input intensity, which are polished but somewhat formulaic
- Limitations and conclusion sections, where the rhetoric is very smooth and generalized relative to the specificity of the empirical setup
Assessed epistemic impact: Likely limited, provided the numerical results, data construction, and code are authentic and reproducible. The main risk is not fabrication but rhetorical smoothing that slightly overextends interpretation beyond the empirical identification strategy.
- Fraud / paper-mill screening
I do not see strong paper-mill indicators. The manuscript fits a recognizable empirical agricultural-economics/panel-data paradigm, uses a coherent model, reports diagnostics, and includes plausible limitations and references . The main concern is not fraud but inferential overreach from proxy-based measurement. No obvious contradiction suggests fabrication on the face of the text alone.
Recommendations
-
Reframe the manuscript explicitly as a cross-country descriptive panel study of associations, not as evidence about hemp-specific fertilizer requirements or producer decision mechanisms. Keep the strongest causal and behavioral language out of the discussion unless new variables are introduced.
-
Add a fuller data transparency subsection reporting:
Item Recommendation Sample construction Report country counts, country-year counts, and losses from zero/missing-value exclusions for each specification Variables Define fertilizer units precisely and explain harmonization steps Replication Provide code, cleaning scripts, and model scripts in a repository Panel coverage Show which countries drive the FE estimates and for how many years -
Tighten the sustainability claims. The paper can reasonably say that output expansion was not strongly linked to higher estimated fertilizer intensity in this dataset, but it should not imply environmental efficiency without hemp-specific agronomic or environmental indicators.
-
Add one or more robustness checks if feasible, such as:
Check Rationale Lagged fertilizer specification Helps assess timing and reduce simultaneity concerns Alternative transformations or semi-log forms Tests sensitivity to zero-exclusion and log specification Influential-country sensitivity analysis Assesses whether large producers dominate the FE estimates Time-trend interactions or subperiod analysis Tests whether associations differ across regulatory phases -
Move some interpretive speculation into a short final paragraph labeled as hypotheses for future research. This would preserve the paper’s interesting contextual insights while protecting it from criticism that the evidence is being stretched.
-
Improve presentation quality by cleaning equation formatting, standardizing capitalization, and ensuring that all tables and figures are fully self-explanatory.
Overall, I judge the manuscript as methodologically competent and publishable in principle after revision, but it currently needs more disciplined alignment between what the data identify and what the discussion claims.
XSci (AI review)
Summary
The manuscript asks to what extent fertilizer consumption is associated with hemp production and productivity across countries, after accounting for cultivated area and country-specific characteristics. The authors assemble an unbalanced country-year panel for 2013–2023 covering "True hemp, raw or retted" and "Hempseed", with production, harvested area and yield, merged with national fertilizer consumption. Log-log models are estimated by pooled OLS and country-and-year fixed effects, with diagnostics for heteroskedasticity, serial correlation and cross-sectional dependence, and country-clustered standard errors. The reported results (Table 2, N = 126) give an area elasticity of 1.075 (pooled) and 1.139 (FE) and a fertilizer elasticity of 0.190 (pooled) and 0.545 (FE). The authors conclude that hemp expansion over the period was predominantly extensive, driven by land allocation rather than input intensification.
Contribution and Significance
The contribution is a cross-country panel description of hemp production dynamics in 2013–2023 and an attempt to separate scale effects from input effects within countries. The data assembly is the real novelty: the hemp economics literature is dominated by enterprise budgets and single-country case studies, and the cited agronomic reviews (Amaducci et al. 2015; Schluttenhofer & Yuan 2017; Sebastian et al. 2023; Višković et al. 2023) are not cross-country econometric work.
The econometric contribution, however, matters less than the manuscript claims. The central fertilizer result is estimated from aggregate national fertilizer consumption, which carries no information about hemp specifically, and the area result is close to an accounting identity, since production equals area times yield and yield is itself constructed as production divided by area (Section 2.1). What remains that is genuinely new relative to prior work is the descriptive mapping of national production regimes in Section 2.2 and Figure 1, not an identified input–output relationship.
Strengths
- The panel covering 2013–2023 for two distinct hemp items, harmonised to ISO3 codes, fills a real gap; no comparable cross-country econometric treatment of hemp appears in the literature.
- Section 2.2 and Figure 1 produce a useful taxonomy of national production regimes: persistent producers (China, France), expansion followed by plateau or contraction (Poland, Spain), and intermittent or marginal engagement (Lithuania, Greece, Hungary, Japan).
- Section 3.3 reports diagnostics with their statistics rather than asserting them (Wooldridge χ² = 13.63, p = 0.001; Pesaran CD z = −1.30, p = 0.194) and adopts country-clustered standard errors in response.
- Sections 2.1 and 4.1 state the non-crop-specific nature of the fertilizer measure openly and refrain from causal language in the Results.
Recommendation
Reject for Agriculture (MDPI). The central claim — that fertilizer consumption is associated with hemp production and productivity — cannot be supported with the data used, and no revision within the manuscript's own approach can repair it, because the required crop-specific input data do not enter the study. The remaining advance, that hemp output tracks harvested area, is close to definitional and is too small to carry the paper. Compounding this, the yield specification that the abstract, Section 2.1 and Section 2.3 promise is never reported, yet the Discussion draws conclusions from it.
Major Concerns
-
The fertilizer variable cannot bear the interpretation placed on it (Section 2.1; Table 2; Section 4, paragraphs 5 and 7). The regressor is total national fertilizer consumption. Within a country, after absorbing year effects, that series moves with the national agricultural cycle and with the size of the whole cropping sector; hemp is a marginal crop in every country in the sample. The fixed-effects estimate of 0.545 implies that a one percent rise in a country's entire fertilizer use accompanies a 0.55 percent rise in hemp output, which is not credible as an input response and is better read as hemp output co-moving with aggregate agricultural conditions. The authors' defence in Section 2.1 — that the measurement error "is likely to attenuate the estimated coefficient towards zero", so the elasticities are "conservative associations" — holds only for classical measurement error. Here the proxy is a different variable, correlated with country agricultural scale and with omitted drivers of hemp area, so the sign of the bias is not known and the estimate cannot be treated as a lower bound. The sustainability reading in Section 4 — "supports the characterization of hemp as a crop capable of expanding without heavy reliance on increased chemical input intensity" — is therefore not licensed by the evidence. Resolving this requires hemp-specific or at least hemp-area-weighted input data; absent that, the fertilizer claim should be withdrawn rather than qualified.
-
Analyses the manuscript relies on are never reported (Sections 2.3, 2.4, 3.1–3.3). Section 2.3 specifies a yield equation and Section 2.1 says the derived yield measure "is used in robustness specifications", but no yield regression appears anywhere; the only yield evidence is the pooled bivariate scatter of Figure 3. The Discussion nonetheless states that the production interpretation "is reinforced by the weaker and less robust relationship observed between fertilizer use and yield", and the closing paragraph of Section 4 claims fertilizer is "positively but modestly associated with hemp production and productivity". Similarly, Section 2.4 states that random effects models are estimated and that "model selection between the random effects and fixed effects specifications is guided by the Hausman test", but neither the random effects estimates nor a Hausman statistic is reported. Claims resting on unreported output cannot be checked.
-
The area elasticity is near-definitional and is over-interpreted (Section 2.1; Table 2; Section 4, paragraph 3). Because production equals area times yield, and yield is computed as production over area when not directly reported, regressing log production on log area recovers an identity plus the area–yield covariance; a coefficient near one and a pooled R² of 0.869 follow mechanically. For the same reason the yield equation in Section 2.3 is the production equation with the area coefficient restricted to one, and so is not an independent robustness check. The Discussion devotes a paragraph to the elasticity "exceeding one" and to "more than proportional increases in output", but with 1.139 and a standard error of 0.140 the deviation from unity is 0.99 standard errors and is not statistically distinguishable from one.
-
The study cannot be reproduced as described (Section 2.1; Table 1; Table 2 note). The data source is never named — Section 2.1 says only "international agricultural statistics sources" — and the fertilizer variable has no stated unit, definition or source: Table 1 reports a mean of 1,259 and a maximum of 27,444 with no unit, leaving the reader to infer whether this is nutrient tonnes, product tonnes, or an intensity measure. The estimation sample is given only as N = 126, with no country count and no country list, even though inference rests on standard errors clustered at the country level; Section 2.2 names about twelve countries, and cluster-robust inference with so few clusters is anti-conservative, so the reported p < 0.01 on fertilizer is not secure without a wild cluster bootstrap or an equivalent correction.
Minor Comments
- Section 3.3 describes "BP = 25.26, p = 0.014" as strongly rejecting homoskedasticity; p = 0.014 does not warrant "strongly".
- The text immediately after the production equation in Section 2.3 is garbled and truncated ("The term $\alpha|\alpha|_{i}|"), and α_i, δ_t and ε_it are never defined.
- The Table 2 note states that year fixed effects are included in both columns, so the first column is not pooled OLS and should be relabelled.
- Section 3.2 and the Conclusion rest a fibre-versus-seed contrast on "descriptive inspection" of hempseed, but no hempseed table or figure is presented.
- Yield is described as "measured in tonnes per hectare" in Section 2.1 but reported as kg/ha in Table 1.
- Excluding zero and missing values for the log transformation (Section 2.1) drops exactly the entry and exit episodes that Section 2.2 and Figure 1 emphasise; the resulting selection on the dependent variable deserves explicit discussion.
- No methodological references are cited for the tests used — Hausman (1978, Econometrica 46(6), 1251–1271), Breusch & Pagan (1979, Econometrica 47(5), 1287–1294), Wooldridge (2002, Econometric Analysis of Cross Section and Panel Data, MIT Press) and the Pesaran CD test (Pesaran 2015, Econometric Reviews 34(6–10), 1089–1117).
- The simultaneity of input and output decisions, the oldest problem in production-function estimation (Marschak & Andrews 1944, Econometrica 12(3), 143–205), is not discussed; with only six references, all from the agronomy and sustainability literature, the economics literature is absent.
- The captions of Figures 1–3 give neither axis units, nor the number of countries plotted, nor the specification of the fitted lines.
- Report the fertilizer variable as an intensity (for example kg per hectare of cropland) alongside the national total, and show whether the results survive.
- Report robustness to dropping China, which the text identifies as the dominant producer and which will heavily influence both the pooled fit and the within-country estimates.
- The compiled panel and the R code should be deposited rather than made "available from the author upon reasonable request", since the underlying sources are public.