pith. sign in

arxiv: 2606.28017 · v1 · pith:HGJV54YGnew · submitted 2026-06-26 · 💰 econ.EM · stat.ME

A Toolkit for the Study of Treatment-Effect Discontinuities

Alessandro Baldi Antognini , Paolo Verme This is my paper

Pith reviewed 2026-06-29 02:03 UTC · model grok-4.3

classification 💰 econ.EM stat.ME
keywords treatment effects curvedistributional treatment effectstreatment effect discontinuitiescausal forestscrossing point asymptoticshorizontal discontinuity analysisvertical discontinuity analysisPROGRESA
0
0 comments X

The pith

A toolkit locates and tests sign changes in the treatment effects curve using horizontal and vertical discontinuity analyses plus adapted crossing-point asymptotics.

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

The paper establishes methods for identifying where marginal distributional treatment effects flip sign along the treatment effects curve. It introduces horizontal discontinuity analysis that uses causal forests to compare regions with opposite-signed effects and vertical discontinuity analysis focused on sign-switch points. The work also adapts crossing-point asymptotics to pinpoint zero crossings of the curve and applies a bias-corrected Wald statistic to test whether the local slope is non-tangential. These tools are shown to work on both synthetic examples and real data from Mexico's PROGRESA program, extending standard distributional treatment effect analyses beyond averages.

Core claim

The central claim is that treatment-effect discontinuities, defined as points where marginal distributional effects change sign, can be studied with a framework that combines Horizontal Discontinuity Analysis comparing groups in opposite-signed regions via causal forests, Vertical Discontinuity Analysis examining sign-switch points, and adapted crossing-point asymptotics that locate zero crossings of the Treatment Effects Curve and test non-tangentiality of its local slope with a bias-corrected Wald statistic.

What carries the argument

The Treatment Effects Curve (TEC), which traces marginal distributional effects as a function of treatment intensity, together with the Horizontal Discontinuity Analysis (HDA) and Vertical Discontinuity Analysis (VDA) procedures and the adapted crossing-point asymptotics that locate and test its zero crossings.

If this is right

  • Analysts can systematically identify regions of positive versus negative treatment effects rather than relying on average effects alone.
  • Sign-switch points along the treatment effects curve become locatable and testable for non-tangentiality.
  • The horizontal and vertical analyses complement existing distributional treatment effect tools by focusing on discontinuities.
  • The full workflow can be applied directly to observational program data such as PROGRESA to reveal where effects change direction.
  • Bias-corrected asymptotics provide a formal test that distinguishes genuine crossings from tangential behavior.

Where Pith is reading between the lines

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

  • The same HDA and VDA steps could be applied to other conditional cash transfer programs to map where benefits turn negative at different quantiles.
  • Combining the zero-crossing locator with existing regression discontinuity designs might allow researchers to treat estimated discontinuity points as new running variables.
  • If the methods scale to high-dimensional covariates, they could help diagnose treatment-effect heterogeneity in settings with many potential moderators.
  • Extensions that relax the single-curve assumption might reveal multiple crossing points within the same dataset.

Load-bearing premise

The treatment effects curve exists as a well-defined object that can be consistently estimated so its zero-crossing points can be located and tested without extra regularity conditions that would invalidate the discontinuity detection.

What would settle it

Controlled synthetic data in which the true treatment effects curve has no zero crossings yet the HDA, VDA, or bias-corrected Wald test reports a discontinuity, or data with known crossings where the methods fail to locate them.

Figures

Figures reproduced from arXiv: 2606.28017 by Alessandro Baldi Antognini, Paolo Verme.

Figure 1
Figure 1. Figure 1: Cumulative distribution functions (CDFs) and Treatment Effects Curves (TECs) for the synthetic data, under two comparisons. Top row reports the pre-post comparison. Panel (a) shows the pre- and post-treatment CDFs and panel (b) the corresponding TEC ∆(x) = Fpre(x) − Fpost(x). Bottom row reports the control-treated comparison among post-treatment units. Panel (c) shows the control (non-treated) and treated … view at source ↗
Figure 2
Figure 2. Figure 2: Treatment effects across covariates by group, causal forest estimates (synthetic data). (a) RDD plot at the first estimated crossing point (b) RDD plot at the second estimated crossing point [PITH_FULL_IMAGE:figures/full_fig_p043_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: RDD plots at the two estimated TEC crossing points (synthetic data). 43 [PITH_FULL_IMAGE:figures/full_fig_p043_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Treatment Effects Curves for PROGRESA, monthly food consumption per adult equiv￾alent, treatment versus control villages. Panels (a) and (b) eligible households, panels (c) and (d) ineligible households in the same villages. Shaded regions are pointwise 95% confidence intervals from a village-level cluster bootstrap (999 replications), dotted lines are uniform (sup-norm) 95% bands. Positive values indicate… view at source ↗
read the original abstract

This paper provides a toolkit for the study of distributional treatment effects (DTEs) focused on treatment-effect discontinuities defined as points where marginal distributional effects change sign. Building on the Treatment Effects Curve (TEC, Verme, 2010), the paper makes three contributions. First, we propose a methodological framework comprising a Horizontal Discontinuity Analysis (HDA) comparing groups in regions of opposite-signed effects using causal forests, and a Vertical Discontinuity Analysis (VDA) examining sign-switch points. Second, we adapt crossing-point asymptotics to locate where a TEC crosses zero and to test the non-tangentiality of its local slope with a bias-corrected Wald statistic. Third, we illustrate the full workflow on synthetic data and add a diagnostic application to Mexico's PROGRESA data. The paper shows how these contributions complement and expand existing instruments for DTE analyses.

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.

Referee Report

2 major / 2 minor

Summary. The paper develops a methodological toolkit for analyzing treatment-effect discontinuities within distributional treatment effects (DTEs), defined as points where marginal distributional effects change sign. Building on the Treatment Effects Curve (TEC) from Verme (2010), it introduces Horizontal Discontinuity Analysis (HDA) that employs causal forests to compare groups in regions of opposite-signed effects, Vertical Discontinuity Analysis (VDA) focused on sign-switch points, and an adaptation of crossing-point asymptotics to locate TEC zero-crossings and test non-tangentiality via a bias-corrected Wald statistic. The framework is illustrated on synthetic data and diagnostically applied to Mexico's PROGRESA dataset, positioning these tools as complements to existing DTE instruments.

Significance. If the proposed HDA, VDA, and adapted asymptotics are shown to be valid under the maintained assumptions, the toolkit could meaningfully expand the set of tools available for DTE analysis in economics by enabling systematic detection and testing of sign changes in treatment effects. The integration of causal forests for horizontal comparisons and the bias-corrected Wald test for non-tangentiality represent practical extensions, and the PROGRESA application provides a concrete empirical illustration. The synthetic-data workflow demonstration is a strength for reproducibility and pedagogical value.

major comments (2)
  1. [Abstract (methodological framework description)] The central claim that the adapted crossing-point asymptotics and bias-corrected Wald statistic can locate TEC zero-crossings and test non-tangentiality without additional regularity conditions that would invalidate detection rests on the unverified assumption that the TEC from Verme (2010) is consistently estimable as a well-defined object; no derivation or simulation evidence is supplied to confirm this holds under the HDA/VDA setup.
  2. [Abstract (contributions 1 and 2)] The independence of the new HDA (causal-forest comparison of opposite-signed regions) and VDA (sign-switch analysis) from the fitted TEC cannot be assessed, raising the risk that the discontinuity detection is circular; this is load-bearing because the paper positions these as extensions that expand existing DTE instruments.
minor comments (2)
  1. [Abstract] The abstract mentions a 'full workflow' on synthetic data and a PROGRESA diagnostic but provides no details on performance metrics, robustness checks, or comparison to alternative DTE methods; adding these would improve clarity.
  2. Notation for the TEC, HDA, and VDA is introduced without explicit definitions or references to the equations in Verme (2010) that are being extended; this could be clarified in the main text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. Below we address each major comment directly, providing clarifications on the role of the TEC and the relationship of HDA/VDA to it. Where the manuscript requires additional detail or evidence, we indicate the planned revisions.

read point-by-point responses

  1. Referee: The central claim that the adapted crossing-point asymptotics and bias-corrected Wald statistic can locate TEC zero-crossings and test non-tangentiality without additional regularity conditions that would invalidate detection rests on the unverified assumption that the TEC from Verme (2010) is consistently estimable as a well-defined object; no derivation or simulation evidence is supplied to confirm this holds under the HDA/VDA setup.

    Authors: The TEC and its consistency properties are taken directly from Verme (2010), which establishes the object under standard regularity conditions for distributional treatment effects. Our crossing-point asymptotics and bias-corrected Wald statistic are adaptations that inherit those conditions; the paper does not claim to relax them. The synthetic-data section illustrates the full workflow, including zero-crossing location, but does not contain a dedicated Monte Carlo study isolating the TEC estimator under the HDA/VDA partitioning. We agree that an explicit verification would strengthen the claim and will add a targeted simulation exercise in the revised manuscript to confirm consistency of the TEC estimator when the sign regions are subsequently used for HDA and VDA. revision: yes

  2. Referee: The independence of the new HDA (causal-forest comparison of opposite-signed regions) and VDA (sign-switch analysis) from the fitted TEC cannot be assessed, raising the risk that the discontinuity detection is circular; this is load-bearing because the paper positions these as extensions that expand existing DTE instruments.

    Authors: HDA and VDA are constructed to take the sign pattern of the TEC as given for region definition, but the subsequent steps—causal-forest estimation of group differences in HDA and direct examination of sign-switch locations in VDA—are separate estimation procedures that do not re-use the TEC functional form or its fitted values. The TEC serves only as an initial classifier of regions; the causal-forest comparisons and sign-switch tests are performed on the raw data within those regions. We acknowledge that the current manuscript does not explicitly demonstrate this separation in a dedicated subsection, which could leave the independence open to question. We will revise the methods section to include a clearer statement and a small illustrative diagram showing the information flow, thereby making the non-circular nature assessable. revision: partial

Circularity Check

1 steps flagged

Self-citation on base TEC object; new HDA/VDA methods retain independent content

specific steps
  1. self citation load bearing [Abstract]
    "Building on the Treatment Effects Curve (TEC, Verme, 2010), the paper makes three contributions. First, we propose a methodological framework comprising a Horizontal Discontinuity Analysis (HDA) comparing groups in regions of opposite-signed effects using causal forests, and a Vertical Discontinuity Analysis (VDA) examining sign-switch points."

    The central object (TEC) whose discontinuities are studied is supplied solely by prior work of co-author Verme; the HDA/VDA and Wald statistic are then defined relative to sign changes in that object. While the new procedures add content, the premise that the TEC is a well-defined, consistently estimable curve whose zero-crossings admit the proposed asymptotics rests on the self-citation without independent derivation or external benchmark in the present paper.

full rationale

The paper explicitly builds its framework on the TEC from Verme (2010) by the co-author, which supplies the core object whose sign changes are analyzed. However, the specific contributions (causal-forest HDA, VDA sign-switch examination, and bias-corrected Wald crossing-point asymptotics) introduce new methodological steps that do not reduce to the prior definition by construction. No fitted-input-as-prediction or self-definitional loops appear in the provided abstract and description; the self-citation is load-bearing for the object but not for the novel tools themselves. This yields moderate circularity without forcing the entire result.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no explicit free parameters, axioms, or invented entities are stated. The framework implicitly rests on standard causal-inference regularity conditions and the existence of the prior TEC object.

pith-pipeline@v0.9.1-grok · 5677 in / 1193 out tokens · 29084 ms · 2026-06-29T02:03:43.481626+00:00 · methodology

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

44 extracted references · 6 canonical work pages

  1. [1]

    Bootstrap tests for distributional treatment effects in instrumental variable models

    Alberto Abadie. Bootstrap tests for distributional treatment effects in instrumental variable models. Journal of the American Statistical Association, 2002

    2002

  2. [2]

    Instrumental variables estimates of the effect of subsidized training on the quantiles of trainee earnings

    Alberto Abadie, Joshua Angrist, and Guido Imbens. Instrumental variables estimates of the effect of subsidized training on the quantiles of trainee earnings. Econometrica : journal of the Econometric Society, 70 0 (1): 0 91--117, 2002

    2002

  3. [3]

    Indirect effects of an aid program: How do cash transfers affect ineligibles' consumption? American Economic Review, 99 0 (1): 0 486--508, 2009

    Manuela Angelucci and Giacomo De Giorgi. Indirect effects of an aid program: How do cash transfers affect ineligibles' consumption? American Economic Review, 99 0 (1): 0 486--508, 2009

    2009

  4. [4]

    Microcredit impacts: Evidence from a randomized microcredit program placement experiment by Compartamos Banco

    Manuela Angelucci, Dean Karlan, and Jonathan Zinman. Microcredit impacts: Evidence from a randomized microcredit program placement experiment by Compartamos Banco . American Economic Journal: Applied Economics, 7 0 (1): 0 151--182, 2015

    2015

  5. [5]

    Generalized random forests.The Annals of Statistics, 47(2):1148–1178, 2019

    Susan Athey, Julie Tibshirani, and Stefan Wager. Generalized random forests. Annals of Statistics, 47 0 (2): 0 1148--1178, 2019. doi:10.1214/18-AOS1709

    work page doi:10.1214/18-aos1709 2019

  6. [6]

    Atkinson

    Anthony B. Atkinson. On the measurement of inequality. Journal of Economic Theory, 2 0 (3): 0 244--263, 1970

    1970

  7. [7]

    The miracle of microfinance? Evidence from a randomized evaluation

    Abhijit Banerjee, Esther Duflo, Rachel Glennerster, and Cynthia Kinnan. The miracle of microfinance? Evidence from a randomized evaluation. American Economic Journal: Applied Economics, 7 0 (1): 0 22--53, 2015

    2015

  8. [8]

    Barrett and Stephen G

    Garry F. Barrett and Stephen G. Donald. Consistent tests for stochastic dominance. Econometrica, 71 0 (1): 0 71--104, 2003

    2003

  9. [9]

    Ready for boarding? The effects of a boarding school for disadvantaged students

    Luc Behaghel, Cl \'e ment de Chaisemartin, and Marc Gurgand. Ready for boarding? The effects of a boarding school for disadvantaged students. American Economic Journal: Applied Economics, 9 0 (1): 0 140--164, 2017. doi:10.1257/app.20150090

    work page doi:10.1257/app.20150090 2017

  10. [10]

    Bishop, John P

    John A. Bishop, John P. Formby, and Paul D. Thistle. Statistical inference, income distribution, and social welfare. Econometrica, 57 0 (5): 0 1057--1065, 1989

    1989

  11. [11]

    Bitler, Jonah B

    Marianne P. Bitler, Jonah B. Gelbach, and Hilary W. Hoynes. What mean impacts miss: Distributional effects of welfare reform experiments. American Economic Review, 96 0 (4): 0 988--1012, 2006

    2006

  12. [12]

    Bitler, Jonah B

    Marianne P. Bitler, Jonah B. Gelbach, and Hilary W. Hoynes. Distributional impacts of the Self-Sufficiency Project . Journal of Public Economics, 92 0 (3-4): 0 748--765, 2008

    2008

  13. [13]

    Quantile treatment effects in difference in differences models under dependence restrictions and with only two time periods

    Brantly Callaway, Tong Li, and Tatsushi Oka. Quantile treatment effects in difference in differences models under dependence restrictions and with only two time periods. Journal of Econometrics, 206 0 (2): 0 395--413, 2018. doi:10.1016/j.jeconom.2018.06.008

    work page doi:10.1016/j.jeconom.2018.06.008 2018

  14. [14]

    The effects of 401(k) participation on the wealth distribution: An instrumental quantile regression analysis

    Victor Chernozhukov and Christian Hansen. The effects of 401(k) participation on the wealth distribution: An instrumental quantile regression analysis. Review of Economics and Statistics, 86 0 (3): 0 735--751, 2004

    2004

  15. [15]

    Inference on counterfactual distributions

    Victor Chernozhukov, Iv \'a n Fern \'a ndez-Val, and Blaise Melly. Inference on counterfactual distributions. Econometrica, 81 0 (6): 0 2205--2268, 2013

    2013

  16. [16]

    Generic machine learning inference on heterogeneous treatment effects in randomized experiments, with an application to immunization in india

    Victor Chernozhukov, Mert Demirer, Esther Duflo, and Iv \'a n Fern \'a ndez-Val . Generic machine learning inference on heterogeneous treatment effects in randomized experiments, with an application to immunization in india. NBER Working Paper 24678, National Bureau of Economic Research, 2018

    2018

  17. [17]

    Estimating the impact of microcredit on those who take it up: Evidence from a randomized experiment in Morocco

    Bruno Cr \'e pon, Florencia Devoto, Esther Duflo, and William Parient \'e . Estimating the impact of microcredit on those who take it up: Evidence from a randomized experiment in Morocco . American Economic Journal: Applied Economics, 7 0 (1): 0 123--150, 2015

    2015

  18. [18]

    Statistical inference for stochastic dominance and for the measurement of poverty and inequality

    Russell Davidson and Jean-Yves Duclos. Statistical inference for stochastic dominance and for the measurement of poverty and inequality. Econometrica, 68 0 (6): 0 1435--1464, 2000

    2000

  19. [19]

    Heterogeneous impacts in PROGRESA

    Habiba Djebbari and Jeffrey Smith. Heterogeneous impacts in PROGRESA . Journal of Econometrics, 145 0 (1-2): 0 64--80, 2008

    2008

  20. [20]

    Kjell A. Doksum. Empirical probability plots and statistical inference for nonlinear models in the two-sample case. The Annals of Statistics, 2 0 (2): 0 267--277, 1974

    1974

  21. [21]

    Sahn, and Stephen D

    Jean-Yves Duclos, David E. Sahn, and Stephen D. Younger. Robust multidimensional poverty comparisons. Economic Journal, 116 0 (514): 0 943--968, 2006

    2006

  22. [22]

    Peer effects, teacher incentives, and the impact of tracking: Evidence from a randomized evaluation in Kenya

    Esther Duflo, Pascaline Dupas, and Michael Kremer. Peer effects, teacher incentives, and the impact of tracking: Evidence from a randomized evaluation in Kenya . American Economic Review, 101 0 (5): 0 1739--1774, 2011. doi:10.1257/aer.101.5.1739

    work page doi:10.1257/aer.101.5.1739 2011

  23. [23]

    Ferraro and Michael K

    Paul J. Ferraro and Michael K. Price. Using nonpecuniary strategies to influence behavior: Evidence from a large-scale field experiment. Review of Economics and Statistics, 95 0 (1): 0 64--73, 2013. doi:10.1162/REST_a_00344

    work page doi:10.1162/rest_a_00344 2013

  24. [24]

    Efficient semiparametric estimation of quantile treatment effects

    Sergio Firpo. Efficient semiparametric estimation of quantile treatment effects. Econometrica, 75 0 (1): 0 259--276, 2007

    2007

  25. [25]

    Fortin, and Thomas Lemieux

    Sergio Firpo, Nicole M. Fortin, and Thomas Lemieux. Decomposing wage distributions using recentered influence function regressions. Econometrics, 6 0 (2): 0 28, 2018

    2018

  26. [26]

    Local linear forests

    Rina Friedberg, Julie Tibshirani, Susan Athey, and Stefan Wager. Local linear forests. Journal of Computational and Graphical Statistics, 30 0 (2): 0 503--517, 2021. doi:10.1080/10618600.2020.1831930

    work page doi:10.1080/10618600.2020.1831930 2021

  27. [27]

    Nonparametric regression for treatment evaluation

    Markus Fr \"o lich. Nonparametric regression for treatment evaluation. Econometrica, 78 0 (2): 0 491--518, 2010

    2010

  28. [28]

    Josef Hadar and William R. Russell. Rules for ordering uncertain prospects. American Economic Review, 59 0 (1): 0 25--34, 1969

    1969

  29. [29]

    The efficiency analysis of choices involving risk

    Giora Hanoch and Haim Levy. The efficiency analysis of choices involving risk. Review of Economic Studies, 36 0 (3): 0 335--346, 1969

    1969

  30. [30]

    D. L. Hawkins and S. C. Kochar. Inference for the crossing point of two continuous CDFs . Statistics & Probability Letters, 11 0 (2): 0 167--170, 1991

    1991

  31. [31]

    Stochastic dominance and expected utility: Survey and analysis

    Haim Levy. Stochastic dominance and expected utility: Survey and analysis. Management Science, 38 0 (4): 0 555--593, 1992

    1992

  32. [32]

    A nonparametric test for treatment effect heterogeneity

    Markus Maier. A nonparametric test for treatment effect heterogeneity. Econometric Reviews, 30 0 (5): 0 635--660, 2011

    2011

  33. [33]

    Understanding the average impact of microcredit expansions: A Bayesian hierarchical analysis of seven randomized experiments

    Rachael Meager. Understanding the average impact of microcredit expansions: A Bayesian hierarchical analysis of seven randomized experiments. American Economic Journal: Applied Economics, 11 0 (1): 0 57--91, 2019

    2019

  34. [34]

    Aggregating distributional treatment effects: A Bayesian hierarchical analysis of the microcredit literature

    Rachael Meager. Aggregating distributional treatment effects: A Bayesian hierarchical analysis of the microcredit literature. American Economic Review, 112 0 (6): 0 1818--1847, 2022

    2022

  35. [35]

    Quantile regression forests

    Nicolai Meinshausen. Quantile regression forests. Journal of Machine Learning Research, 7 0 (35): 0 983--999, 2006

    2006

  36. [36]

    Orthogonal random forest for causal inference

    Miruna Oprescu, Vasilis Syrgkanis, and Zhiwei Steven Wu. Orthogonal random forest for causal inference. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors, Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 4932--4941. PMLR, 2019

    2019

  37. [37]

    On estimation of a probability density function and mode

    Emanuel Parzen. On estimation of a probability density function and mode. Annals of Mathematical Statistics, 33 0 (3): 0 1065--1076, 1962

    1962

  38. [38]

    Nonparametric estimation of distributional policy effects

    Christoph Rothe. Nonparametric estimation of distributional policy effects. Journal of Econometrics, 155 0 (1): 0 56--70, 2010

    2010

  39. [39]

    Stiglitz

    Michael Rothschild and Joseph E. Stiglitz. Increasing risk: I. a definition. Journal of Economic Theory, 2 0 (3): 0 225--243, 1970

    1970

  40. [40]

    S. J. Sheather and M. C. Jones. A reliable data-based bandwidth selection method for kernel density estimation. Journal of the Royal Statistical Society, Series B, 53 0 (3): 0 683--690, 1991

    1991

  41. [41]

    Shorrocks

    Anthony F. Shorrocks. Ranking income distributions. Economica, 50 0 (197): 0 3--17, 1983

    1983

  42. [42]

    Stochastic dominance, poverty and the treatment effect curve

    Paolo Verme. Stochastic dominance, poverty and the treatment effect curve. Economics Bulletin, 30 0 (1): 0 365--373, 2010. URL https://ideas.repec.org/a/ebl/ecbull/eb-09-00688.html

    2010

  43. [43]

    Estimation and inference of heterogeneous treatment effects using random forests

    Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 114 0 (525): 0 324--337, 2019

    2019

  44. [44]

    M. P. Wand and M. C. Jones. Kernel Smoothing. Chapman and Hall, London, 1995

    1995