arxiv: 2603.28470 · v2 · submitted 2026-03-30 · 💰 econ.EM · stat.ME

Recognition: 2 theorem links

· Lean Theorem

Counterfactual Density Effects and the German East--West Income Gap

Georg Keilbar , Sonja Greven

Authors on Pith no claims yet

Pith reviewed 2026-05-14 00:44 UTC · model grok-4.3

classification 💰 econ.EM stat.ME

keywords causal inferencecounterfactual densitiesEast-West Germanyincome gapdistributional effectsBayes Hilbert spacefunctional regressionOaxaca-Blinder decomposition

0 comments

The pith

A density-based causal method decomposes the full German East-West income gap into distribution and covariate effects.

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

The paper develops a framework for causal inference that operates on entire outcome densities rather than averages alone. It decomposes observed density differences between treated and control groups into a distribution effect, which changes the conditional density while fixing covariates at the treated distribution, and a covariate effect, which changes the covariate distribution. Estimation proceeds by embedding conditional densities in a Bayes Hilbert space and fitting a functional additive regression model that respects the density constraints of non-negativity and unit integration. The approach is applied to post-reunification wage data for East and West Germans, capturing features such as probability mass at zero that mean-based methods overlook. Sympathetic readers would care because it supplies a more complete account of how group differences arise across the entire outcome range.

Core claim

The central claim is that counterfactual density effects, obtained by altering either the conditional density of the outcome or the distribution of covariates, together account for observed density discrepancies between East and West Germans and carry causal interpretations under unconfoundedness and overlap. The Bayes Hilbert space representation allows the densities to be treated as elements of a linear space while preserving their probabilistic properties, and the functional additive regression model estimates the required conditional densities flexibly.

What carries the argument

The Bayes Hilbert space representation of conditional densities together with functional additive regression, which enables decomposition of density differences into distribution and covariate effects while maintaining non-negativity and integration-to-one.

If this is right

The two counterfactual effects fully account for the observed density discrepancy between groups.
Both effects receive causal interpretations under unconfoundedness and overlap.
The method identifies causal impacts on any feature of the outcome distribution, including boundary probabilities such as mass at zero.
Applied to wages, it reveals distributional nuances in the East-West gap that average comparisons miss.

Where Pith is reading between the lines

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

Analysts of other group disparities could use the same decomposition to determine whether covariate shifts or conditional-response changes dominate.
The framework could be extended to panel data to track how the relative size of the two effects evolves over time.
Policy design might target the larger of the two effects once they are separately identified.

Load-bearing premise

The classical unconfoundedness and overlap assumptions hold, so the constructed counterfactual densities correspond to causal rather than merely associational effects.

What would settle it

In a setting with known random assignment, the estimated counterfactual densities would differ substantially from the observed post-treatment densities, or sensitivity analysis would show that modest violations of unconfoundedness reverse the sign or magnitude of the reported effects.

read the original abstract

We propose a novel framework for conducting causal inference based on counterfactual densities. While the current paradigm of causal inference is mostly focused on estimating average treatment effects (ATEs), which restricts the analysis to the first moment of the outcome variable, our density-based approach is able to detect causal effects based on general distributional characteristics. Following the Oaxaca-Blinder decomposition approach, we consider two types of counterfactual density effects that together explain observed discrepancies between the densities of the treated and control group. First, the distribution effect is the counterfactual effect of changing the conditional density of the control group to that of the treatment group, while keeping the covariates fixed at the treatment group distribution. Second, the covariate effect represents the effect of a hypothetical change in the covariate distribution. Both effects have a causal interpretation under the classical unconfoundedness and overlap assumptions. Methodologically, our approach is based on analyzing the conditional densities as elements of a Bayes Hilbert space, which preserves the non-negativity and integration-to-one constraints. We specify a flexible functional additive regression model estimating the conditional densities. We apply our method to analyze the German East--West income gap, i.e., the observed differences in wages between East Germans and West Germans. While most of the existing studies focus on the average differences and neglect other distributional characteristics, our density-based approach is suited to detect all nuances of the counterfactual distributions, including differences in probability masses at zero.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper sets up a clean density version of Oaxaca-Blinder but the abstract gives no results to judge whether it changes anything in practice.

read the letter

The main advance is a decomposition of density differences into a distribution effect (swap conditional densities, hold covariates fixed) and a covariate effect, both given causal meaning under standard unconfoundedness and overlap. They embed the conditional densities in a Bayes Hilbert space and estimate them with functional additive regression so the objects stay non-negative and integrate to one. That part is technically tidy and directly extends the mean-only logic without adding new identifying assumptions. The East-West income application is a sensible test case because the wage distributions differ in ways that means alone would miss, such as mass near zero. The construction itself looks internally consistent; the functional model is just an estimation device that targets the same counterfactual objects as the classical decomposition. The clear limitation is that the text supplies no fitted densities, no standard errors, no robustness checks, and no comparison to a plain mean decomposition. Without those pieces it is impossible to tell whether the extra machinery delivers different or more reliable conclusions on this data. The unconfoundedness assumption is invoked but not tested or relaxed. This is for labor economists already working with distributional outcomes who want a formal way to move beyond averages. If the full paper contains stable estimates and clear diagnostics it would be worth refereeing; otherwise it stays a methodological note without demonstrated payoff.

Referee Report

2 major / 3 minor

Summary. The paper develops a framework for causal inference on counterfactual densities by embedding conditional densities in the Bayes Hilbert space and using functional additive regression. It defines two counterfactual effects: the distribution effect (changing conditional density while fixing covariates) and the covariate effect (changing covariate distribution). These have causal interpretations under unconfoundedness and overlap. The method is applied to the German East-West income gap to analyze full distributional differences, including mass at zero.

Significance. If validated, this provides a tool to go beyond average effects in causal analysis, useful for understanding heterogeneous impacts on distributions. The preservation of density constraints is a technical strength. The application to East-West gap illustrates potential for labor market studies, but the significance depends on whether the empirical findings reveal new insights not captured by mean-based methods.

major comments (2)

[§5 (Empirical Application)] §5 (Empirical Application): The reported counterfactual densities for East and West Germans show differences at zero income, but the manuscript does not include formal statistical tests or bootstrap confidence bands for these density estimates, making it hard to determine if the detected effects are statistically significant.
[§3.2 (Functional Additive Regression)] §3.2 (Functional Additive Regression): The model specification assumes an additive structure for the log-density effects; however, no diagnostic or test is provided to assess whether this additivity holds in the German income data, which could bias the decomposition if interactions are present.

minor comments (3)

The abstract mentions 'differences in probability masses at zero' but the main text should explicitly define how zero-income observations are handled in the density estimation.
[§2] §2: Clarify the exact definition of the Bayes Hilbert space inner product used for the embedding, perhaps with a brief example calculation.
Add a reference to recent work on distributional causal inference, such as papers using Wasserstein distances or other functional approaches.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and the recommendation of minor revision. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [§5 (Empirical Application)] The reported counterfactual densities for East and West Germans show differences at zero income, but the manuscript does not include formal statistical tests or bootstrap confidence bands for these density estimates, making it hard to determine if the detected effects are statistically significant.

Authors: We agree that providing measures of statistical uncertainty strengthens the empirical application. In the revised manuscript, we will add bootstrap confidence bands for the estimated counterfactual densities (including the pointwise bands at zero income) using a nonparametric bootstrap procedure that accounts for the functional nature of the densities and the two-stage estimation process. revision: yes
Referee: [§3.2 (Functional Additive Regression)] The model specification assumes an additive structure for the log-density effects; however, no diagnostic or test is provided to assess whether this additivity holds in the German income data, which could bias the decomposition if interactions are present.

Authors: We acknowledge the importance of validating the additivity assumption. In the revision, we will include a diagnostic check by comparing the additive model fit to a specification that allows selected interactions (via tensor-product terms) using cross-validation on the functional regression criterion. We will report the results and discuss any implications for the decomposition if substantial departures from additivity are detected. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external causal assumptions and standard decomposition

full rationale

The paper's central claims rest on the classical unconfoundedness and overlap assumptions for causal interpretation of the two counterfactual density effects (distribution effect and covariate effect). These are invoked directly from the literature without reduction to fitted parameters or self-citation chains. The Bayes Hilbert space embedding and functional additive regression are presented as estimation tools that enforce non-negativity and integration-to-one by construction, but they target the same counterfactual objects defined via the Oaxaca-Blinder logic and do not alter the identifying content. No equations reduce a claimed prediction to an input fit by definition, and no uniqueness theorem or ansatz is smuggled via self-citation. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on two standard causal assumptions plus the choice of Bayes Hilbert space representation; no free parameters or invented entities are described in the abstract.

axioms (2)

domain assumption Unconfoundedness assumption
Invoked to give counterfactual density effects a causal interpretation.
domain assumption Overlap assumption
Invoked to give counterfactual density effects a causal interpretation.

pith-pipeline@v0.9.0 · 5547 in / 1150 out tokens · 81162 ms · 2026-05-14T00:44:58.030996+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean and IndisputableMonolith/Foundation/BranchSelection.lean Jcost symmetry under inversion; RCLCombiner_isCoupling_iff (c ≠ 0 forces bilinear branch) echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

we consider ratios, fY⟨1,1⟩(y)/fY⟨0,0⟩(y) = fY⟨1,1⟩(y)/fY⟨0,1⟩(y) × fY⟨0,1⟩(y)/fY⟨0,0⟩(y). The motivation for the use of a multiplicative decomposition is twofold. First, for the estimation of conditional densities we rely on the use of Bayes Hilbert spaces, which are suitable spaces for density functions. These are vector spaces in which addition corresponds to multiplication and subtraction corresponds to taking ratios
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean embed_add (multiplicative homomorphism of LogicNat orbit) echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

fi = ⊕j=1^J hj(xi) … clr(fi) = … (b(xi) ⊗ b̃T)^⊤ θ

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.