{"id":"db8cf164-fc0e-4152-9413-52f04d248633","arxiv_id":"2508.20114","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"Across 2,903 cities, population size tracks total building area but essentially not average building height, implying vertical development adds little capacity for residents.","lead":"A study of 2,903 cities finds that population scales mostly with building footprint area and barely with average building height, suggesting that vertical development contributes little to population accommodation. If true, this challenges the assumption that high-rises are an efficient way to house growing urban populations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Z-scored regression coefficients are read as Cobb-Douglas elasticities: the reported α_h≈0, α_A<1, and σ are standardized betas, not the exponents of Eq. (3), so the central 'height does not matter' claim is unquantified.","rationale":"Agreement: The reader's weakest assumption identifies the same issue. I confirm it is load-bearing. The paper's Eq. (3) defines α_A and α_h as exponents; Eq. (4) calls them partial derivatives. But §2.1 says all variables are standardized before regression. In a standardized log-log OLS, the coefficients are β_A = α_A·s_A/s_P and β_h = α_h·s_h/s_P. Since the paper never converts back, every reported α is actually a standardized beta. The t-statistics of the zero test are invariant under rescaling, so a statement 'not significantly different from zero' would survive; but the paper makes stronger quantitative claims: α_h≈0, α_A<1, and σ regimes. Those magnitudes can change arbitrarily with within-country variance ratios. For example, if within-country variation in ln h is much smaller than in ln P, β_h will be near zero even if the raw elasticity α_h is meaningful. This directly undermines the abstract's conclusion that vertical extent does not contribute. The substitution rate in Eq. (5) is also distorted because it is the ratio of variance-scaled coefficients. Therefore the central claim as reported is not supported. A re-analysis with raw variables is a straightforward and decisive check. I would keep REJECT rather than CONDITIONAL because the reported numbers are not what they are claimed to be; however, the concern is fixable and the empirical finding may survive.","tokens_in":12697,"tokens_out":6547,"duration_ms":73025,"concrete_test":"Using the posted data/code, refit Eq. (3) for each of the 42 countries without any standardization: regress ln P on ln A and ln h. Report raw α_A, α_h with 90% CIs. Then check two quantities: (i) In countries where Fig. 2c shows α_h≈0, does the raw α_h CI contain 0 and is |α_h|<0.1? If not, the height conclusion flips. (ii) For a few countries, compute β_h from the raw fit and compare with the paper's reported α_h; this verifies whether they are standardized betas. Also compute s_h/s_P for countries with β_h≈0; if this ratio is small, the standardized value understates raw elasticity. If raw α_h remains centered on zero and raw α_A in (0,1) for the same countries, the standardization concern is resolved and the verdict could be revised upward; otherwise the reported exponents must be corrected and conclusions re-evaluated.","verdict_should_be":"REJECT","load_bearing_attack":"The paper's central estimates come from a log-log OLS in which, per §2.1, 'we standardize all variables before regression,' and the resulting coefficients are then presented as α_A and α_h in Eq. (3)/(4) and used to compute σ in Eq. (5). If ln P, ln A, and ln h are z-scored before fitting, the estimated coefficients are standardized betas β_A, β_h, not the partial elasticities α_A=∂lnP/∂lnA and α_h=∂lnP/∂lnh. The two are related by α_A=β_A·s_P/s_A and α_h=β_h·s_P/s_h, where s are within-country standard deviations of the logged variables. No conversion back to raw elasticities is reported. This is consequential: β_h≈0 can coexist with a non-negligible raw α_h whenever s_h≪s_P, e.g., if average height varies little across cities while population varies a lot. Thus the headline finding 'α_h≈0' may reflect low cross-city variance in height, not the absence of a structural relationship. Moreover, the reported 0<α_A<1 pattern and the substitution rate σ=−α_h/α_A are variance-scaled; σ computed from standardized coefficients equals −(β_h/β_A)(s_A/s_h), which differs from the raw trade-off. Since the absolute sizes of the coefficients and their comparison across countries drive the paper's conclusions, the central claim as stated is not supported by the reported statistics.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a Cobb-Douglas scaling model P ∼ A^αA h^αh linking population to total building footprint area and average building height, and fits it to 2903 Functional Urban Areas in 42 countries using the 3D-GloBFP dataset. The authors report that αA is positive and typically below one, while αh is approximately zero for most countries, which they interpret as evidence that vertical development does not significantly contribute to population accommodation. They further compute a substitution rate σ = −αh/αA, relate αA and αh to country-level metrics (Global North/South, ζ-exponent, GDP, urbanization) via Lasso regression, and run a building-type classification robustness check (Germany, Philippines, Spain).","tokens_in":13119,"tokens_out":6197,"duration_ms":73940,"significance":"The question of whether cities accommodate population through horizontal expansion or vertical density is of broad urban-science and policy relevance, and the paper brings a new global dataset to bear on it. The authors have made data and code available, and the cross-country design is more comprehensive than earlier single-country or single-city studies. If the quantitative results were correct, the finding that height contributes little to population accommodation would challenge a common policy intuition. However, the central estimates are currently not the quantities they are claimed to be because of standardization; this must be resolved before the paper's conclusions can be evaluated. The paper's strongest asset is the dataset and the multivariate formulation; its current quantitative evidence for the headline claim is not valid.","major_comments":[{"comment":"The paper states 'we standardize all variables before regression' and then reports the coefficients as αA and αh, defined in Eq. (4) as ∂lnP/∂lnA and ∂lnP/∂lnh. These are not the same. If lnP, lnA, and lnh are z-scored before fitting, the OLS coefficients are standardized betas βA and βh, related to the raw partial elasticities by αA = βA·sP/sA and αh = βh·sP/sh. No conversion is reported. Consequently, the headline αh≈0 may simply reflect a small cross-city variance of lnh relative to lnP within countries, rather than an absence of structural association. The claim that building height does not contribute to population accommodation is therefore not supported by the reported statistics. Please re-estimate in raw log-log space or report converted elasticities together with the within-country standard deviations.","section":"§2.1, Eqs. (3)-(4)"},{"comment":"The substitution rate σ = −αh/αA is used to classify countries into complementary, no-interaction, and substitution regimes. When computed from standardized coefficients, the quantity is actually −βh/βA · sA/sh, not the slope of the constant-population isoquant derived in Eq. (14). Thus the regimes in Fig. 3 and the associated interpretation of how area and height trade off are variance-scaled artifacts of the standardization. The isoquant plots (Figs. 3d-e) should be based on the raw model, and σ must be recomputed from the raw elasticities.","section":"§2.3, Eq. (5), Fig. 3"},{"comment":"The derivation of the negative αA–αh relationship imposes consistency between the Cobb-Douglas model and the bivariate scaling laws A∼P^βA and V∼P^βV, which are fitted on the same data. Eq. (13) is therefore an algebraic consistency condition, not an independent theoretical constraint. The negative correlation in Fig. 2d is partly a mathematical consequence of this identity and of the estimation procedure, so it should not be presented as independent evidence of a volume-constrained trade-off. If the standardization issue is corrected, this relationship may change and should be re-examined.","section":"§4.2, Eq. (13), Fig. 2d"},{"comment":"The data-cleaning procedure excludes countries with fewer than 30 FUAs, applies a population threshold cut-off, and then excludes countries with fewer than 10 remaining FUAs. These choices are not justified, and truncating the city-size distribution can bias the estimated scaling exponents. Because the country-level αA and αh are the paper's main outputs, the sensitivity of the results to the threshold choices should be reported (e.g., varying the minimum number of FUAs and the population cutoff).","section":"§4.1.1"}],"minor_comments":[{"comment":"'Variance Inflation Factor' should be 'the variance inflation factor'. Also, the text refers to Supplementary Figure S1 for all countries, but only example countries are shown in the main text; please clarify what the supplementary figure contains.","section":"§2.1"},{"comment":"In the final paragraph of the Discussion, 'approch' should be 'approach'.","section":"Discussion"},{"comment":"The caption has grammatical issues: 'the white horizontal line represent' should be 'represents', and 'with the color denote' should be 'with the color denoting'.","section":"Fig. 4 caption"},{"comment":"Reference [40] is cited without a year or volume/page information; please provide the full publication details.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The standardization issue is the main technical problem: the reported αA and αh are not the elasticities defined in Eq. (3), so the headline conclusion is currently unsupported. However, this is fixable by re-running the regressions in raw log space or converting standardized betas to raw elasticities using the reported data. The referee report focuses on this point; if the authors provide the raw elasticities and the conclusion changes, the paper should be reconsidered at that point. The circularity in Eq. (13) should also be addressed by framing it as a consistency condition rather than independent theoretical support."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it puts a genuinely new global dataset to work: 2,903 FUAs across 42 countries, with building footprint area and average height from 3D-GloBFP, and it makes data and code available. That is real value. Second, the central regression is run on z-scored variables, and the resulting coefficients are presented as the Cobb-Douglas elasticities α_A and α_h. That is a plain methodological error, and it is load-bearing.\n\nWhat the paper does well: the cross-country scope is new, the building-classification robustness check is a nice touch, and the Cobb-Douglas framing is a reasonable way to think about horizontal and vertical dimensions jointly, building on the authors' prior work. The negative correlation between α_A and α_h (RMA slope −0.67) is interesting, though it is partly a mathematical consequence of the constraint they derive from bivariate scaling fits on the same data, not an independent confirmation.\n\nWhere it falls down: if ln P, ln A, and ln h are each standardized before OLS, the estimated coefficients are standardized betas, related to the true partial elasticities by α_A = β_A·s_P/s_A and α_h = β_h·s_P/s_h. The paper never converts back. So a small β_h can hide a non-trivial α_h whenever cross-city variation in average height is small relative to variation in population — which is exactly plausible for this kind of data. The headline result, that vertical development contributes almost nothing to population accommodation, is therefore unquantified. The same problem affects the substitution rate σ, which is computed from the standardized coefficients. The data-cleaning thresholds (countries with <30 FUAs, then additional population cut-offs) add selection concerns, but they are secondary; the standardization issue alone is enough to undercut the central claim.\n\nIf the authors re-run the regressions on unstandardized logs and report the raw elasticities, the result might survive. I would not bet on it either way from the present numbers, and I would want to see the raw standard deviations before believing any magnitude. The paper deserves a serious referee because the dataset and question are worth engaging with, but it needs major revision — an editor should send it out, and a referee should insist on seeing the unstandardized results before taking the conclusions seriously.\n\nFor a reading group, it's a useful case study in why standardization matters in log-log regressions. I would bring it up, but I would not cite it in my own work as it stands.","headline":"Valuable new dataset and a sharp question, but the central claim rests on standardized regression coefficients being read as elasticities, so 'height doesn't matter' is not currently supported.","tokens_in":13565,"tokens_out":1744,"would_cite":false,"duration_ms":22937,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Building height shows almost no association with city population size once built area is counted.","keywords":["urban scaling","building height","Cobb-Douglas model","horizontal expansion","vertical development","functional urban areas","population accommodation","cross-sectional analysis"],"falsifier":"Re-run the country-level Cobb-Douglas regression without standardizing, i.e., on raw log(P), log(A), and log(h); if for a country like Poland the raw height coefficient is clearly positive and stable, the claim that height contributes nothing fails for that country. Alternatively, a within-city panel that follows cities after large additions of tall residential buildings would contradict the cross-sectional null if population grows with added vertical floor area.","tokens_in":12643,"feed_emoji":"🏙️","tokens_out":7954,"duration_ms":88161,"temperature":0.7,"pith_summary":"This paper tries to establish that, when city population is related at the same time to the total footprint of its buildings and to their average height, the height term contributes almost nothing. Fitting the two-factor multiplicative relation P ~ A^αA h^αh separately for 42 national urban systems (2,903 functional urban areas in 2015), the authors find αA is positive and usually below 1, while αh is near zero in most countries and even negative in a few. The consequence, if true, is that today's cities accommodate people mainly by spreading out horizontally, and building taller has not, on average, been an effective way to increase population capacity. A reader should care because this challenges a standard intuition behind high-rise urban policy and attaches a concrete number to how strongly each dimension tracks population.","feed_headline":"Building height barely explains city population size","feed_subtitle":"Across 2,903 cities in 42 countries, only horizontal building area reliably tracks population.","key_machinery":"The Cobb-Douglas production form P ∼ A^αA h^αh—a two-factor multiplicative power-law relation—estimated by ordinary least squares on log-transformed, standardized variables is the central object. Its exponents are read as marginal products: how much population changes with a 1% change in total building area, holding average height fixed, and vice versa. The ratio σ = −αh/αA converts the pair of exponents into the slope of a constant-population isoquant, giving three regimes (complementary, independent, substitutive). A geometric identity V = A·h ties the model to the one-dimensional scaling laws and yields the predicted negative relationship between αA and αh.","core_discovery":"The paper's central claim is that the Cobb-Douglas relation P ~ A^αA h^αh, fit city by city within each country, separates horizontal and vertical contributions to population in a way the one-dimensional scaling laws cannot. For most countries the vertical exponent αh is statistically indistinguishable from zero, so taller average building height does not track larger population once area is accounted for; the horizontal exponent αA is positive and typically between 0 and 1, meaning population grows less than proportionally with built footprint. A few countries (Egypt, Morocco, Bangladesh, the Philippines) show negative αh, and some (Germany) show positive αh. The paper also reports a negati","pith_inferences":["Because the regressions are run on standardized variables, the 'height contributes nothing' headline is partly a statement about within-country variance in average height; where all cities have similar heights, αh cannot be large even if vertical space matters at the margin. Re-fitting on raw logs or computing elasticities at the mean would tell how much of αh≈0 is real and how much is a scaling a","The cross-section compares cities at one time, so it cannot distinguish whether population follows height or height follows population; a longitudinal test on cities that add substantial residential tower stock would help decide.","Using average height hides the distribution of building heights. It could be that the top of the height distribution (a few very tall buildings) matters little while moderate mid-rise density matters a lot; decomposing height into floors or per-capita floor area would be a natural extension.","The country-level association between αA and the population-concentration exponent ζ is ecological: it does not say that concentrated countries should expand horizontally, only that their current stock lines up that way. Testing within countries over time would separate national hierarchy from local land-use decisions."],"forward_implications":["If the cross-sectional result is right, vertical development should not be justified chiefly by its population capacity; whatever high-rises deliver has to be measured in other outcomes.","Horizontal built-up area is the dimension that tracks population, but sublinearly (αA < 1), so each added square meter of footprint is associated with proportionally fewer additional residents.","In countries where population is concentrated in a few large cities (high ζ), horizontal expansion is the dimension tied to population; those urban systems may need land-use and density policies different from more evenly distributed countries.","Countries with positive αh, such as Germany, already constrain horizontal growth; the pattern suggests height can contribute where policy forces compactness, while height restrictions elsewhere may push growth into horizontal expansion.","The same P–A–h framework can be applied to other urban quantities (e.g., energy use, emissions, economic output), giving a systematic way to separate horizontal and vertical contributions."],"supporting_citations":[{"why":"Supplies the global 3D building footprint and height data for all 2,903 cities, the empirical basis of the analysis.","marker":"[37]"},{"why":"Defines Functional Urban Areas, the city units to which population and building data are aggregated.","marker":"[61]"},{"why":"Extends and validates the metropolitan-area delineation used for the 2015 city sample.","marker":"[62]"},{"why":"Reviews and formalizes the fundamental allometry A ~ P^βA that the paper generalizes to three variables.","marker":"[31]"},{"why":"Provides the urban scaling framework from which the Cobb-Douglas extension is built.","marker":"[32]"},{"why":"Offers the earlier height-population scaling h ~ P^βh with βh≈1/6, the empirical baseline the paper shows fails generically.","marker":"[36]"},{"why":"Supplies the theoretical reconciliation of horizontal and vertical urban growth that motivates Eq. (3).","marker":"[40]"},{"why":"Provides the ζ-exponent method for city-size distributions used in the country-level external analysis.","marker":"[63]"}],"fun_headline_variants":["Tall buildings don't track city population size","Height barely matters for city population capacity","Horizontal sprawl, not skyscrapers, predicts city residents","Most cities grow outward, not upward, for population","Building height adds little to city population fit"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The result that height does not matter hinges on treating the coefficients of standardized (z-scored) log regressions as structural elasticities; if height varies little across a country's cities, standardization can drive αh toward zero even when vertical space would house people at the margin.","fun_headline_variants_meta":{"raw":{"variants":["Tall buildings don't track city population size","Height barely matters for city population capacity","Horizontal sprawl, not skyscrapers, predicts city residents","Most cities grow outward, not upward, for population","Building height adds little to city population fit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000257,"raw_usage":{"total_tokens":1388,"prompt_tokens":688,"completion_tokens":700,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":432,"completion_tokens_details":{"reasoning_tokens":627}},"tokens_in":432,"tokens_out":700,"duration_ms":8056,"temperature":1.0,"reasoning_tokens":627,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T17:59:44.817004+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the country-level Cobb-Douglas regression without standardizing, i.e., on raw log(P), log(A), and log(h); if for a country like Poland the raw height coefficient is clearly positive and stable, the claim that height contributes nothing fails for that country. Alternatively, a within-city panel that follows cities after large additions of tall residential buildings would contradict the cross-sectional null if population grows with added vertical floor area.","supporting_citations":[{"cited_title":"Earth System Science Data 2024, 1–28 (2024) https://doi.org/10.5194/essd-2024-217","cited_arxiv_id":null,"evidence_quote":"Supplies the global 3D building footprint and height data for all 2,903 cities, the empirical basis of the analysis."},{"cited_title":"Technical report, JRC Technical Report (2019)","cited_arxiv_id":null,"evidence_quote":"Defines Functional Urban Areas, the city units to which population and building data are aggregated."},{"cited_title":"Delineation and population trends","cited_arxiv_id":null,"evidence_quote":"Extends and validates the metropolitan-area delineation used for the 2015 city sample."},{"cited_title":"Environment and Planning B https://doi.org/10.1177/23998083251332324","cited_arxiv_id":null,"evidence_quote":"Supplies the theoretical reconciliation of horizontal and vertical urban growth that motivates Eq. (3)."},{"cited_title":"PLoS ONE 16(1), 0245771 (2021) https://doi.org/ 10.1371/journal.pone.0245771 22","cited_arxiv_id":null,"evidence_quote":"Provides the ζ-exponent method for city-size distributions used in the country-level external analysis."}],"review_version":1}