Pith. sign in

REVIEW 3 major objections 5 minor 217 references

The paper proves that the counterfactual composition for a treated unit is the weighted geometric mean of donor shares when a random-utility factor model and a convex-hull condition hold, turning synthetic controls into a strictly compositi

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 19:19 UTC pith:FEWLF7TF

load-bearing objection Legitimate SCM-for-compositions method, but the Pennsylvania application is underidentified: more donor weights than pre-treatment effective observations, so the 60 pp effect is not pinned down. the 3 major comments →

arxiv 2607.16991 v1 pith:FEWLF7TF submitted 2026-07-18 econ.EM stat.ME

Compositional Synthetic Controls

Onil Boussim This is my paper
classification econ.EM stat.ME
keywords compositional datasynthetic controlAitchison geometrylog-odds transformationrandom utility modelcounterfactual inferenceelectricity generation mixtreatment effect
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper extends the synthetic control method to outcomes that are vectors of shares, such as an electricity generation mix. It starts from a multinomial logit model: the log-odds of each category against a baseline are exactly the relative systematic utilities. Imposing an interactive fixed-effects structure on those utilities, and assuming the treated unit's factor loadings lie in the convex hull of donor loadings, the counterfactual log-odds become a convex combination of donor log-odds. In share space, this means the counterfactual is the renormalized weighted geometric mean of donor shares—the Fréchet barycenter under the Aitchison metric. The estimator stacks all log-odds equations into one quadratic program, and the Pennsylvania application finds natural gas's share exceeded its counterfactual by nearly 60 percentage points by 2022.

Core claim

The central claim is Proposition 1: under Assumptions 1–3, the treated unit's counterfactual log-odds equal a convex combination of donor log-odds up to a mean-zero error, so the counterfactual composition is the unique Fréchet barycenter of the donor compositions under the Aitchison metric. Concretely, the counterfactual share vector is the renormalized weighted geometric mean of donor shares, with a single weight vector applied across all categories. This yields a valid composition by construction and identifies the counterfactual in the simplex's own geometry. In the application to Pennsylvania's electricity generation after the Alternative Energy Portfolio Standard, the estimated effect

What carries the argument

The log-odds map ℓ(π) = (log π1/πp, …, log πp−1/πp) and its inverse (the softmax) convert the simplex into Euclidean space, where a linear factor model on relative utilities is the natural specification. Identification is carried by the convex-hull condition (Assumption 3) on the unit-level factor loadings; the estimator then solves a convex quadratic program on the stacked (p−1)T₀ log-odds equations, giving a single weight vector for all categories. The Fréchet barycenter under the Aitchison metric—the renormalized weighted geometric mean—is the closed-form counterfactual that follows from this structure.

Load-bearing premise

That the treated unit's unobserved factor loadings can be written as a convex combination of the donors' loadings (Assumption 3); in the Pennsylvania application this is fragile because no donor state shares Pennsylvania's Marcellus shale endowment, so the post-2004 natural-gas boom may be a unit-specific factor the donor pool cannot span.

What would settle it

Check whether Pennsylvania's pre-treatment log-odds vector lies inside the convex hull of donor log-odds for every pre-treatment year; if any year's point falls outside the hull, Assumption 3 is violated and the counterfactual is extrapolation. Equivalently, re-estimate using only donors that share Pennsylvania's gas-bearing geology; if the 60-point gas gap collapses, the headline effect is identification-driven rather than causal.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Counterfactual compositions are automatically valid: every share is positive and the vector sums to one.
  • A single weight vector applies to all categories, avoiding the incoherent multiple-counterfactual problem of running separate SCMs for each share.
  • The effective sample size is (p−1)T₀, so consistency can be driven by the number of categories as well as the length of the pre-treatment period.
  • The Euclidean SCM objective is a share-weighted distortion of the Aitchison objective (Proposition 3), meaning it systematically down-weights small-share categories; the new estimator treats all log-ratio discrepancies symmetrically.
  • In Pennsylvania, the policy effect is a compositional reallocation: natural gas gains almost 60 percentage points by 2022 while renewables, despite growing in absolute terms, lose on the relative scale—a pattern a single-share analysis would miss.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the log-odds representation extends beyond logit to any invertible link, the method may transfer to other discrete-choice families such as nested logit or GEV with only a change in the transformation.
  • The same machinery applies to vote shares, expenditure shares, employment-sector shares, or any categorical aggregate where the policy question is how mass is reallocated between categories.
  • Where the convex-hull condition fails—for example, a treated unit with a unique resource endowment like Marcellus shale—the 'synthetic' path is extrapolation rather than interpolation; a diagnostic would be to test whether the treated unit's pre-treatment log-odds lie inside the donor hull in every period, not just on average.
  • The permutation-based inference bound (minimum p-value ≈ 1/J) makes the Pennsylvania result (p=0.111) suggestive rather than decisive; conformal inference adapted to the Aitchison metric is a natural extension for exact confidence sets.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes Compositional Synthetic Controls (CSC), an estimator for compositional outcomes on the simplex. It starts from a random utility model with interactive fixed effects on relative systematic utilities, so that log-odds of shares equal latent preference parameters. Under a convex-hull condition on factor loadings (Assumption 3), the treated unit's counterfactual log-odds are a convex combination of donor log-odds (Proposition 1), and the counterfactual composition is the Fréchet barycenter under the Aitchison metric, i.e., the renormalized weighted geometric mean of donor compositions. Weights are estimated by a stacked quadratic program over (p-1)T0 log-odds observations. The method is illustrated with Pennsylvania's electricity generation mix after the 2004 Alternative Energy Portfolio Standard, reporting a natural-gas share gap of about 60 percentage points by 2022. Placebo inference based on Aitchison RMSPE ratios gives a p-value of 0.111.

Significance. If the identification conditions are met, CSC is a principled and computationally simple extension of synthetic controls to compositions. The log-odds map connects aggregate shares to a discrete-choice microfoundation, and the interpretation of the counterfactual as a weighted geometric mean under the Aitchison metric is elegant. The stacking across categories following Sun et al. (2025) is a useful specialization, and the paper provides a clear algorithm and replication package. However, the empirical claims are currently not supported: the application is underidentified, and the consistency proof has a missing rank condition. These issues are load-bearing for the manuscript as written.

major comments (3)
  1. [§6.2, Eq. (18)] The Pennsylvania application has J=42 donors, so J-1=41 weights, and N=(p-1)T0=2*14=28 scalar observations. The stacked design matrix X is 28×41 with rank at most 28, so the quadratic objective in Eq. (18) is flat along at least a 13-dimensional affine subspace of the simplex. The reported weights in Table 1 are one arbitrary solution; the near-perfect pre-treatment fit does not identify w*. Proposition 2's consistency argument requires N→∞ with J fixed, whereas here N=28 and J=42, so it does not apply. Section 7 itself notes that penalization can improve finite-sample performance when the donor pool is large relative to (p-1)T0, exactly this case, but §6 uses no penalty. Consequently the reported 60 pp treatment effect and donor weights are not identified.
  2. [§6.1, Assumption 3 (Eq. 12)] Identification of the counterfactual relies on the treated unit's factor loadings lying in the convex hull of donor loadings. The Pennsylvania donor pool excludes states with comparable AEPS policies and contains no Marcellus-shale state. The post-2004 natural-gas boom in Pennsylvania appears driven by a unit-specific endowment not shared by any donor, making the synthetic gas path an extrapolation rather than an interpolation. The paper does not assess the plausibility of Assumption 3 for this application or provide sensitivity diagnostics. This undermines the causal interpretation of the empirical section.
  3. [Appendix A.1 (Proof of Proposition 2)] The proof asserts that uniqueness of w* in Assumption 3 ensures Q(w)>Q(w*) for all w≠w*. This is not implied unless the stacked factor matrices {Θ_t, Λ_t} over the pre-treatment sample have full rank to detect any discrepancy (Z1 - Σw_jZ_j, μ1 - Σw_jμ_j). If the factor columns are collinear or the covariates are constant, multiple weights can make the deterministic discrepancy zero in-sample even when Eq. (12) fails. A rank condition on the factor loading matrices is missing; the proof as written is incomplete.
minor comments (5)
  1. [Throughout] The main identification result is Proposition 1, but the text refers to 'Theorem 1' in several places (Section 7, Remark 3, proof of Proposition 2). Please renumber consistently.
  2. [Abstract and §6.4] The abstract states that the application 'uncovers a large and persistent compositional shift' without reporting that the placebo p-value is 0.111, which is not significant at conventional levels. The headline should be qualified.
  3. [§3.3, Eq. (14)] Eq. (13) is exact, so the counterfactual composition given the weights is exactly ℓ^{-1}(Σ w*_j ℓ(π_jt)+η). The O_p(||η||) term and the 'first-order expansion' in the proof are unnecessary and slightly misleading.
  4. [§5, Algorithm 1] The zero-share perturbation constant c is a free parameter; the paper should discuss sensitivity of results to c (currently set to 10^-6) and to the placebo screen threshold m.
  5. [Appendix A.1] The independence assumption 'ε_{k,j,t} independent across (k,t)' is strong for compositional data, where categories within a unit may have correlated shocks. A brief discussion or relaxation would be helpful.

Circularity Check

0 steps flagged

No significant circularity: the derivation is a model-based identification argument, not a fitted-parameter prediction.

full rationale

The paper's central derivation is self-contained in the sense required by the circularity analysis. Equation (8) is an identity from the multinomial logit model: given the logit formula, observable log-odds equal relative systematic utilities by construction. This is a modeling assumption, not a renaming of the target result. Proposition 1 then shows that under Assumption 3 (convex hull of factor loadings), the treated unit's counterfactual log-odds are a convex combination of donor log-odds. The proof is algebraic given the factor model and the convex-hull condition; Assumption 3 is an untestable identifying assumption, not a parameter fitted to the outcome being predicted. The Aitchison/Fréchet equivalence in equation (16) is a mathematical correspondence between weighted log-odds averaging and the closed geometric mean, correctly attributed to Aitchison (1982) and Egozcue et al. (2003); presenting this known equivalence as a geometric interpretation is not circular. The weights are estimated using only pre-treatment data in equations (17)-(18), and the post-treatment counterfactual uses those weights with donor post-treatment compositions; this is standard out-of-sample synthetic control prediction, not a fitted input masquerading as a prediction. The paper's finite-sample identification concern in the Pennsylvania application (N=28 effective observations versus J-1=41 weights) is a substantive correctness and identification issue, but it is not a circularity: the reported weights and effects are under-identified, not forced by construction or by self-citation. No load-bearing self-citations appear; the stacking principle of Sun et al. (2025) is an external methodological borrowing, not an unverified self-referential premise. Therefore the derivation chain does not reduce to its own inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 7 axioms · 0 invented entities

The central estimator rests on the logit inversion, a factor model on relative utilities, and the convex-hull condition; no new entities are introduced. The application adds donor-selection and placebo-screen choices that are not free parameters of the estimator.

free parameters (2)
  • Placebo screen threshold m = 5
    Chosen by hand as a multiple of the treated unit's pre-treatment RMSPE to retain donors; affects the placebo set and p-value in §5.
  • Zero-share perturbation c = 1e-6
    Mentioned in Algorithm 1 for relaxing Assumption 1; not actually used because zero-share states are excluded from the application.
axioms (7)
  • domain assumption Multinomial logit with iid Type-I extreme value taste shocks
    §3.1, Eq. (5)–(6): aggregate shares equal the logit choice probabilities; delivers the log-odds = relative utility identity.
  • domain assumption Interactive fixed effects on relative systematic utilities, with factor loadings common across categories
    §3.2, Eq. (10)–(11): the structural justification for a single weight vector across all categories.
  • domain assumption Convex hull condition on observed covariates Z and unobserved loadings μ
    Assumption 3, Eq. (12): needed for Proposition 1 identification; standard in SCM but untestable.
  • domain assumption Positivity of all shares
    Assumption 1: ensures log-odds are well-defined; relaxed only by a perturbation that the application avoids.
  • domain assumption Uniqueness of the convex-hull weights
    Proposition 2: needed for consistency of the QP minimizer; rules out collinear donor configurations.
  • domain assumption Exchangeability of treated and donor units under the sharp null
    §5, Eq. (26): the placebo permutation p-value treats donor placebo ratios as exchangeable with the treated ratio.
  • standard math Standard math: log-odds bijection, Aitchison geometry as Hilbert space, convex QP consistency tools
    Used throughout; these are established results (Aitchison, Egozcue, Newey-McFadden) not introduced by this paper.

pith-pipeline@v1.3.0-alltime-deepseek · 12797 in / 13125 out tokens · 117660 ms · 2026-08-01T19:19:01.381714+00:00 · methodology

0 comments
read the original abstract

This paper develops a synthetic control estimator for compositional outcomes, vectors of shares generated by an underlying categorical process. Derived from a random utility model with interactive fixed effects on relative systematic utilities, the estimator maps compositions to log-odds, where the standard convex hull condition identifies the counterfactual as a convex combination of donor log-odds. Equivalently, it recovers the Fr\'{e}chet barycenter under the Aitchison metric, the canonical geometry of the simplex (the non-linear space of shares) using a single set of weights across all categories. I also developed a placebo inference procedure based on the Aitchison distance. An application to Pennsylvania's electricity generation mix following the Alternative Energy Portfolio Standard uncovers a large and persistent compositional shift: natural gas exceeds its counterfactual by nearly 60 percentage points by 2022, while renewables lose relative ground.

Figures

Figures reproduced from arXiv: 2607.16991 by Onil Boussim.

Figure 1
Figure 1. Figure 1: Generation shares for Pennsylvania and its synthetic control. The synthetic shares are the Fréchet [PITH_FULL_IMAGE:figures/full_fig_p019_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Placebo distribution of post-to-pre Aitchison RMSPE ratios. Pennsylvania (dark) exceeds all but [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Placebo gaps: Aitchison distance between actual and synthetic shares for Pennsylvania (dark) and [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

217 extracted references · 19 linked inside Pith

  1. [1]

    Riemannian ��\^

    Afsari, Bijan , journal=. Riemannian ��\^

  2. [2]

    arXiv preprint arXiv:2505.00331 , year=

    Geodesic synthetic control methods for random objects and functional data , author=. arXiv preprint arXiv:2505.00331 , year=

    arXiv

  3. [3]

    Petersen, Alexander and M. Fr. The Annals of Statistics , volume=. 2019 , publisher=

    2019

  4. [4]

    2023 , publisher=

    Synthetic controls with multiple outcomes: Estimating the effects of non-pharmaceutical interventions in the COVID-19 pandemic , author=. 2023 , publisher=

    2023

  5. [5]

    Journal of the American Statistical Association , volume=

    Smearing estimate: a nonparametric retransformation method , author=. Journal of the American Statistical Association , volume=. 1983 , publisher=

    1983

  6. [6]

    Mathematical geology , volume=

    Isometric logratio transformations for compositional data analysis , author=. Mathematical geology , volume=. 2003 , publisher=

    2003

  7. [7]

    Journal of Machine Learning Research , volume=

    Tangential Wasserstein Projections , author=. Journal of Machine Learning Research , volume=

  8. [8]

    arXiv preprint arXiv:2501.17436 , year=

    Geodesic Difference-in-Differences , author=. arXiv preprint arXiv:2501.17436 , year=

    Pith/arXiv arXiv

  9. [9]

    arXiv preprint arXiv:2406.19604 , year=

    Geodesic Causal Inference , author=. arXiv preprint arXiv:2406.19604 , year=

    Pith/arXiv arXiv

  10. [10]

    arXiv preprint arXiv:2312.15376 , year=

    Geodesic optimal transport regression , author=. arXiv preprint arXiv:2312.15376 , year=

    Pith/arXiv arXiv

  11. [11]

    2020 , publisher=

    An invitation to statistics in Wasserstein space , author=. 2020 , publisher=

    2020

  12. [12]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    The statistical analysis of compositional data , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=. 1982 , publisher=

    1982

  13. [13]

    arXiv preprint arXiv:2510.11659 , year=

    Compositional difference-in-differences for categorical outcomes , author=. arXiv preprint arXiv:2510.11659 , year=

    Pith/arXiv arXiv

  14. [14]

    Review of Economics and Statistics , pages=

    Using multiple outcomes to improve the synthetic control method , author=. Review of Economics and Statistics , pages=. 2025 , publisher=

    2025

  15. [15]

    Journal of economic literature , volume=

    Using synthetic controls: Feasibility, data requirements, and methodological aspects , author=. Journal of economic literature , volume=. 2021 , publisher=

    2021

  16. [16]

    Econometrica , volume=

    Distributional synthetic controls , author=. Econometrica , volume=. 2023 , publisher=

    2023

  17. [17]

    Journal of the American statistical Association , volume=

    Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program , author=. Journal of the American statistical Association , volume=. 2010 , publisher=

    2010

  18. [18]

    American economic review , volume=

    The economic costs of conflict: A case study of the Basque Country , author=. American economic review , volume=. 2003 , publisher=

    2003

  19. [19]

    Journal of the American Statistical Association , volume=

    A penalized synthetic control estimator for disaggregated data , author=. Journal of the American Statistical Association , volume=. 2021 , publisher=

    2021

  20. [20]

    American Economic Journal: Applied Economics , volume=

    Minimum wages and the distribution of family incomes , author=. American Economic Journal: Applied Economics , volume=. 2019 , publisher=

    2019

  21. [21]

    2021 , publisher=

    Topics in optimal transportation , author=. 2021 , publisher=

    2021

  22. [22]

    Journal of Machine Learning Research , volume=

    Network regression with graph Laplacians , author=. Journal of Machine Learning Research , volume=

  23. [23]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Regression for compositional data by using distributions defined on the hypersphere , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2011 , publisher=

    2011

  24. [24]

    The Annals of Applied Statistics , pages=

    Non-Euclidean statistics for covariance matrices, with applications to diffusion tensor imaging , author=. The Annals of Applied Statistics , pages=. 2009 , publisher=

    2009

  25. [25]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Causal inference on distribution functions , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2023 , publisher=

    2023

  26. [26]

    Annals of mathematics , pages=

    Constrained steepest descent in the 2-Wasserstein metric , author=. Annals of mathematics , pages=. 2003 , publisher=

    2003

  27. [27]

    Energy Economics , volume=

    The impact of climate policy on fossil fuel consumption: Evidence from the Regional Greenhouse Gas Initiative (RGGI) , author=. Energy Economics , volume=. 2021 , publisher=

    2021

  28. [28]

    Review of Economic Studies , volume=

    Changes in social network structure in response to exposure to formal credit markets , author=. Review of Economic Studies , volume=. 2024 , publisher=

    2024

  29. [29]

    2019 , institution =

    Alberto Abadie , title =. 2019 , institution =

    2019

  30. [30]

    American Journal of Political Science , volume=

    Comparative politics and the synthetic control method , author=. American Journal of Political Science , volume=. 2015 , publisher=

    2015

  31. [31]

    Annales de l'Institut Henri Poincar

    Geodesic PCA in the Wasserstein space by convex PCA , author=. Annales de l'Institut Henri Poincar. 2017 , organization=

    2017

  32. [32]

    2009 , publisher =

    Shape and shape theory , author =. 2009 , publisher =

    2009

  33. [33]

    SIAM Journal on Mathematical Analysis , volume =

    Barycenters in the Wasserstein space , author =. SIAM Journal on Mathematical Analysis , volume =. 2011 , publisher =

    2011

  34. [34]

    Econometrica , volume=

    Synthetic control as online linear regression , author=. Econometrica , volume=. 2023 , publisher=

    2023

  35. [35]

    2009 , publisher=

    Optimal transport: old and new , author=. 2009 , publisher=

    2009

  36. [36]

    Progress in Nonlinear Differential Equations and their applications , volume=

    Optimal transport for applied mathematicians , author=. Progress in Nonlinear Differential Equations and their applications , volume=. 2015 , publisher=

    2015

  37. [37]

    2008 , publisher =

    Gradient flows in metric spaces and in the space of probability measures , author =. 2008 , publisher =

    2008

  38. [38]

    Atti Accad

    Gradient flows with metric and differentiable structures, and applications to the Wasserstein space , author=. Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl , volume=

  39. [39]

    Trudy Mat

    A theorem on triangles in a metric space and some of its applications , author=. Trudy Mat. Inst. Steklov. , volume=. 1951 , publisher=

    1951

  40. [40]

    2006 , publisher=

    Infinite Dimensional Analysis: a hitchhiker's guide , author=. 2006 , publisher=

    2006

  41. [41]

    IEEE Transactions on Evolutionary Computation , volume=

    A review on bilevel optimization: From classical to evolutionary approaches and applications , author=. IEEE Transactions on Evolutionary Computation , volume=. 2017 , publisher=

    2017

  42. [42]

    Annals of operations research , volume=

    An overview of bilevel optimization , author=. Annals of operations research , volume=. 2007 , publisher=

    2007

  43. [43]

    2006 , publisher=

    Foundations of modern probability , author=. 2006 , publisher=

    2006

  44. [44]

    ACM Transactions on Graphics (TOG) , volume=

    Spherical averages and applications to spherical splines and interpolation , author=. ACM Transactions on Graphics (TOG) , volume=. 2001 , publisher=

    2001

  45. [45]

    Princeton landmarks in mathematics , author=

    Convex Analysis. Princeton landmarks in mathematics , author=. 1997 , publisher=

    1997

  46. [46]

    Advances in mathematics , volume=

    A convexity principle for interacting gases , author=. Advances in mathematics , volume=. 1997 , publisher=

    1997

  47. [47]

    2003 , publisher=

    Topics in Optimal Transportation , author=. 2003 , publisher=

    2003

  48. [48]

    Journal of political economy , volume=

    Identification and estimation of hedonic models , author=. Journal of political economy , volume=. 2004 , publisher=

    2004

  49. [49]

    Advances in neural information processing systems , pages=

    Sinkhorn distances: Lightspeed computation of optimal transport , author=. Advances in neural information processing systems , pages=

  50. [50]

    International Conference on Machine Learning , pages=

    Fast computation of Wasserstein barycenters , author=. International Conference on Machine Learning , pages=

  51. [51]

    Numerical solution of the optimal transportation problem using the Monge--Amp

    Benamou, Jean-David and Froese, Brittany D and Oberman, Adam M , journal=. Numerical solution of the optimal transportation problem using the Monge--Amp. 2014 , publisher=

    2014

  52. [52]

    Numerische Mathematik , volume=

    A computational fluid mechanics solution to the Monge-Kantorovich mass transfer problem , author=. Numerische Mathematik , volume=. 2000 , publisher=

    2000

  53. [53]

    Foundations and Trends

    Computational optimal transport , author=. Foundations and Trends. 2019 , publisher=

    2019

  54. [54]

    The Annals of Statistics , volume=

    Monge--Kantorovich depth, quantiles, ranks and signs , author=. The Annals of Statistics , volume=. 2017 , publisher=

    2017

  55. [55]

    2013 , publisher=

    Weak convergence and empirical processes: with applications to statistics , author=. 2013 , publisher=

    2013

  56. [56]

    Statistical bounds for entropic optimal transport: sample complexity and the central limit theorem , author=

  57. [57]

    Energy Policy , volume=

    The effect of cap-and-trade on sectoral emissions: Evidence from California , author=. Energy Policy , volume=. 2024 , publisher=

    2024

  58. [58]

    Conditional logit analysis of qualitative choice behavior , author=

  59. [59]

    Modelling the choice of residential location , author=

  60. [60]

    Journal of public economics , volume=

    The measurement of urban travel demand , author=. Journal of public economics , volume=. 1974 , publisher=

    1974

  61. [61]

    The Annals of Probability , volume=

    Central limit theorems for empirical transportation cost in general dimension , author=. The Annals of Probability , volume=. 2019 , publisher=

    2019

  62. [62]

    Proceedings of Machine Learning Research , pages =

    Sample Complexity of Sinkhorn Divergences , author =. Proceedings of Machine Learning Research , pages =. 2019 , editor =

    2019

  63. [63]

    arXiv: 1902.01136 , year=

    Directional differentiability for supremum-type functionals: statistical applications , author=. arXiv: 1902.01136 , year=

    Pith/arXiv arXiv 1902

  64. [64]

    Advances in neural information processing systems , pages=

    Stochastic optimization for large-scale optimal transport , author=. Advances in neural information processing systems , pages=

  65. [65]

    2003 , publisher=

    Sobolev spaces , author=. 2003 , publisher=

    2003

  66. [66]

    Probability Theory and Related Fields , volume=

    On the rate of convergence in Wasserstein distance of the empirical measure , author=. Probability Theory and Related Fields , volume=. 2015 , publisher=

    2015

  67. [67]

    The 22nd International Conference on Artificial Intelligence and Statistics , pages=

    Interpolating between Optimal Transport and MMD using Sinkhorn Divergences , author=. The 22nd International Conference on Artificial Intelligence and Statistics , pages=

  68. [68]

    1994 , publisher=

    Stochastic limit theory: An introduction for econometricians , author=. 1994 , publisher=

    1994

  69. [69]

    Annals of Operations Research , volume=

    Asymptotic analysis of stochastic programs , author=. Annals of Operations Research , volume=. 1991 , publisher=

    1991

  70. [70]

    Probability Theory and Related Fields , volume=

    On nondifferentiable functions and the bootstrap , author=. Probability Theory and Related Fields , volume=. 1993 , publisher=

    1993

  71. [71]

    Journal of Econometrics , volume=

    The numerical delta method , author=. Journal of Econometrics , volume=. 2018 , publisher=

    2018

  72. [72]

    The Review of Economic Studies , volume=

    Inference on directionally differentiable functions , author=. The Review of Economic Studies , volume=. 2018 , publisher=

    2018

  73. [73]

    Handbook of econometrics , volume=

    Large sample estimation and hypothesis testing , author=. Handbook of econometrics , volume=. 1994 , publisher=

    1994

  74. [74]

    International Conference on Artificial Intelligence and Statistics , pages=

    Learning Generative Models with Sinkhorn Divergences , author=. International Conference on Artificial Intelligence and Statistics , pages=

  75. [75]

    Econometrica: Journal of the Econometric Society , pages=

    Uniform convergence in probability and stochastic equicontinuity , author=. Econometrica: Journal of the Econometric Society , pages=. 1991 , publisher=

    1991

  76. [76]

    2019 , note=

    Independent nonlinear component analysis , author=. 2019 , note=

    2019

  77. [77]

    Applied Mathematical Sciences , volume=

    A gradient descent solution to the Monge-Kantorovich problem , author=. Applied Mathematical Sciences , volume=

  78. [78]

    American Economic Journal: Applied Economics , volume=

    Minimum wages and the distribution of family incomes , author=. American Economic Journal: Applied Economics , volume=

  79. [79]

    Journal of Statistical Software , year =

    Nonparametric Econometrics: The np Package , author =. Journal of Statistical Software , year =

  80. [80]

    The Review of Economic Studies , volume=

    Decriminalizing indoor prostitution: Implications for sexual violence and public health , author=. The Review of Economic Studies , volume=. 2017 , publisher=

    2017

Showing first 80 references.