Pith. sign in

REVIEW 2 major objections 5 minor 71 references

Evaluating Program Sequences with Double Machine Learning: An Application to Labor Market Policies

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A wage subsidy beats other Swiss job programs, even with equal duration

desk verdict Competent, honest extension of sequential DML to dynamic policies; worth a serious referee, but the causal assumption needs more than a narrative defense. read the letter →

arxiv 2506.11960 v1 pith:5ADREDEQ submitted 2025-06-13 econ.EM

Fabian Muny This is my paper
classification econ.EM MSC 62P2062D20
keywords doublemachinelearningdynamictreatmenteffectspoliciessequentialassignmentactivelabormarketconfoundingprogramevaluationg-formula
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Most evaluation studies collapse a person's whole program history into a single treatment state, which misrepresents how caseworkers actually assign active labor-market measures: the second program is chosen only after seeing how the first one went. This paper argues that the right counterfactual is a dynamic policy, a rule that assigns the first program and then reassigns a second program only to those who remain unemployed, and that this estimand is identified from observational data under a sequential conditional-independence assumption closely parallel to the static one. It then shows that double machine learning, built on the efficient influence-function score for the average outcome under such policies, estimates these effects flexibly while preserving root-N inference. In 177,856 Swiss unemployment-benefit recipients followed through two three-month periods, the analysis finds that a temporary wage subsidy delivers the most months of employment over 30 months, a ranking that survives matching program duration; it also finds that ignoring dynamic confounding exaggerates the benefits of longer programs and that static counterfactuals can reverse conclusions about program order.

What carries the argument

The load-bearing construct is the dynamic policy $g_2(Y^{d_1}_1) = \big(d_1,\; 1\{Y^{d_1}_1 = 0\}\cdot d_2 + 1\{Y^{d_1}_1 = 1\}\cdot \text{NP}\big)$, a rule that assigns program $d_1$ to everyone in the first period and then continues with $d_2$ only for those whose potential intermediate employment status leaves them unemployed, sending the employed to 'no program.' Identification runs through the iterated conditional expectation $\nu_{g_2}(X_0) = E_{X_1}[\mu_{g_2}(X_1) \mid X_0, D_1 = g_1(V_0)]$, the g-formula, and estimation through the doubly robust score $\Theta^{dy}_{g_2}(W_2) = \nu_{g_2}(X_0) + (\mu_{g_2}(X_1) - \nu_{g_2}(X_0))\frac{1\{D_1 = g_1(V_0)\}}{p_{g_1}(X_0)} + (Y - \mu_{g_2}(X_1))\frac{1\{D_2 = (g_1(V_0), g_2(V_1))\}}{p_{g_2}(X_1, g_1)p_{g_1}(X_0)}$, which adds a reweighted outcome residual for each period and satisfies Neyman orthogonality, so that five-fold cross-fitted random-forest estimates of the outcome regressions and propensity scores still yield root-N inference. Two sequential-DML implementations are used, one with ordinary second-order sample splitting and one with an additional doubly robust bias correction, and they give nearly identical answers in this application.

What would settle it

Using the same data, replace the outcome with a variable that predates the first program, such as earnings in the year before the unemployment spell, and run the full dynamic-policy estimator: if the estimated 'effects' of program sequences on this pre-treatment outcome are statistically significant, the sequential unconfoundedness assumption fails and the claimed program ranking is not credible.

Watch

Extended reading notes

Core claim

On its own terms, the paper's contribution is a demonstration that the machinery built for static treatment sequences extends to dynamic policies with only a modest strengthening of assumptions: conditional on baseline covariates, first-period treatment must be independent not only of final potential outcomes but also of the intermediate potential variables the policy depends on, and second-period treatment must be conditionally independent of the final outcome given the full covariate history. Given that, the average outcome under a dynamic policy is identified by the g-formula $\nu_{g_2}(X_0) = E_{X_1}[\mu_{g_2}(X_1) \mid X_0, D_1 = g_1(V_0)]$, and the paper proves the corresponding augmented inverse-probability-weighted score is Neyman-orthogonal, so cross-fitted DML delivers root-N-consistent, asymptotically normal estimates with random-forest nuisance functions. The empirical payoff is the finding that a temporary wage subsidy beats job-search assistance, training courses, and employment programs in months of employment even when program durations are aligned, that this advantage is largest relative to the shortest programs, and that the choice of counterfactual, static versus dynamic, changes which program orderings look best.

Load-bearing premise

The load-bearing premise is that every factor driving a caseworker's assignment decision, at both the first and the second period, is recorded in the data, so that conditional on observed history, program assignment is as good as random.

Editorial extensions

If this is right

  • Evaluating only first programs or imposing fixed two-period sequences misleads: when the second-period program is set to 'no program,' ignoring dynamic confounding nearly doubles the estimated advantage of a long employment program over a short job-search assistance program, so causal claims about program duration need the dynamic framework.
  • A wage subsidy is the strongest program on average, producing roughly 2 to 4 more months of employment over 30 months than the alternatives depending on the counterfactual, with the margin widening when durations are equalized.
  • For individuals with limited local-language skills, subsidized work experience beats extended training, and repeated wage-subsidy participation across spells loses effectiveness, implying that targeting rather than uniform assignment would be more effective.
  • The method's practical ceiling is data support: the number of possible sequences grows exponentially with the number of periods, which is why the application stops at two periods and four program groups, and larger administrative datasets are what would unlock longer sequences.
  • The paper positions DML-based dynamic-policy estimation as a viable complement to parametric sequential methods such as marginal structural models, since it drops functional-form assumptions while keeping root-N inference.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same 'continue only if eligible' rule transfers directly to clinical treatment sequences, for example continuing or switching a drug based on an intermediate biomarker or side-effect profile, so the dynamic-policy estimand and its DML estimator should apply in health settings where observed assignment already conditions on time-varying clinical status.
  • Because the paper notes that Swiss individuals are randomly assigned to caseworkers, a direct check of the identifying assumption is available: instrument the program assignments with caseworker assignment propensities, or run placebo tests on pre-program outcomes, and confirm that the DML estimates do not move.
  • The near-identical results from the two estimator variants suggest the cross-period product-rate condition relaxed by one of the proposed methods is not the binding constraint in rich administrative data; it would matter more with weaker first-stage predictors or fewer observations per sequence.
  • The 7% trimming of extreme-propensity individuals silently changes the target population, and since the trimmed units are more often from the German-speaking region with longer prior unemployment, the policy conclusions are best read as applying to the common-support subpopulation of program participants rather than all benefit recipients.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper develops and applies a double machine learning (DML) framework for evaluating sequential program assignments, with an explicit focus on dynamic policies whose second-period treatment depends on an intermediate outcome. The identifying assumptions are standard sequential conditional independence and overlap conditions, adapted to dynamic policies, and the estimation section reviews orthogonal score-based DML estimators for dynamic confounding, including those of Bodory et al. (2022) and Bradic et al. (2024). The empirical application uses Swiss administrative data on active labor market programs, defining dynamic policies that continue a program in the second period only if the individual remains unemployed. The main empirical conclusion is that a temporary wage subsidy is the most effective program on average, even after accounting for duration differences. The paper also discusses program-order effects and heterogeneity by language skills and prior program participation.

Significance. If the empirical claims hold, the paper makes a useful contribution by connecting dynamic-policy estimands to recently developed sequential DML estimators, and by demonstrating their use in a policy-relevant setting with rich administrative data. The identification proofs in Section 2 and Appendix B are standard and internally coherent, and the authors provide code and transparent implementation details, which strengthens reproducibility. The central empirical conclusion, however, depends on a strong sequential conditional independence assumption that is defended mainly by narrative argument; the lack of a placebo-outcome or sensitivity analysis is a substantive gap. The paper is likely to be of interest to researchers evaluating programs with sequential assignment, but the empirical headline should be treated cautiously pending additional robustness evidence.

major comments (2)
  1. [§4.5 and Assumption 2.5[a]] The identification of dynamic policies requires, in the first period, conditional independence between D1 and both the final potential outcomes Y^{d2} and the intermediate potential decision variables V^{d1}_1, given X0. In the Swiss application, assignment to the first program is made by caseworkers who have access to information that is only partially recorded, so random assignment to caseworkers does not by itself justify this assumption. The paper's defense in §4.5 that the additional restrictiveness 'appears to be unproblematic' is a verbal argument, not an empirical one. Because the sign of the bias from unmeasured confounding is unknown, the headline ranking of wage subsidies relative to other programs in Table 2, Panel D could be overstated. I recommend adding a placebo-outcome test using pre-treatment outcomes, a sensitivity analysis that perturbs the propensity score or outcome model, or a formal partial-identification exercise. This is load-bearing for the empirical conclusion.
  2. [§4.6, cross-fitting and hyperparameter tuning] The DML theory invoked in Section 3 requires that nuisance functions be estimated on a sample independent of the observations used to evaluate the orthogonal score. In §4.6, random forest hyperparameters are tuned on the full sample using FLAML, which includes all cross-fitting folds. This breaks the sample-splitting independence condition that underlies the validity of the cross-fitted estimator and the standard errors reported in Tables 2 and 3. Tuning should be performed within each training fold, or the authors should provide a theoretical justification that full-sample tuning does not affect the asymptotic results. The same concern applies to the propensity-score trimming rule, which is applied once to the full sample before cross-fitting and could similarly induce dependence between nuisance estimates and the evaluation sample.
minor comments (5)
  1. [Definition 2.2] The notation g2 is used both for the full dynamic policy pair and for the second-period component of that pair, which creates confusion in the statement of Assumption 2.5 and Theorem 2.3. Please disambiguate, for example by writing the full policy as g = (g1, g2).
  2. [§4.4, treatment definition] The paper assigns individuals who participate in multiple programs within a period to the longest program, but it does not report how common such within-period multiple participation is. Since this aggregation is acknowledged in the text as an assumption, a descriptive statistic or robustness check excluding or reclassifying such individuals would help assess its impact.
  3. [§4.4 and Appendix F] The handling of individuals with 'Other programs' (OP) in the second period is not fully transparent: the text says they remain in the sample for first-period modeling but are not included in the analysis. It would be clearer to state explicitly how they are treated in the estimation of the propensity scores and outcome models for the policies of interest.
  4. [Table 3] Some entries in Table 3 have very large standard errors, notably 5.54 for the JA-vs-EP comparison under a static policy. These estimates are not informative and should be interpreted cautiously or possibly suppressed with a note about sparse support for the corresponding sequences.
  5. [Figure 3] The labels 'UE no program' and 'Not UE' in the bottom row of Figure 3 are unclear. Please rephrase to indicate exactly which unemployment-history group is being compared.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the identification theorems and DML estimators follow from stated assumptions and external methodological literature; the sole self-citation is for data documentation and is not load-bearing.

full rationale

The paper's derivation chain is self-contained at the level of claims: estimands are defined in terms of potential outcomes, identification is derived from explicit conditional independence and overlap assumptions (Assumptions 2.2–2.5, Theorems 2.1–2.4), and the sequential DML estimators are taken from external work by Bodory et al. (2022) and Bradic et al. (2024) and extended to dynamic policies. The empirical wage-subsidy ranking in Section 4.7 is the output of these estimators applied to administrative data, not a fitted constant or a parameter that was preset to produce that ranking. The only self-citation is Mascolo et al. (2024), a companion data paper used for institutional background and data details; it is not the source of the identification result or of the empirical conclusion. The unverifiable sequential conditional independence assumption (Assumption 2.5) is a substantive identification concern, not a circularity, because the paper's theorems do not assume the effect they estimate. Hyperparameter tuning on the full sample is a statistical practice issue and does not make the predictions algebraic consequences of the fitted values. Overall, the paper does not reduce by construction to its own inputs.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new theoretical entities, particles, forces, or conserved quantities. Its empirical estimates depend on a small set of design parameters, most notably the two-period structure, the three-month period length, the FLAML tuning, and the propensity trimming threshold, plus the standard unconfoundedness and overlap assumptions.

free parameters (5)
  • Length of each period = 3 months
    Chosen in a data-driven way to maximize second-period program starts and covariate changes; it determines the treatment states and all sequences.
  • Number of periods T = 2
    Restricted to two periods because sample sizes for sequences with T > 2 proved insufficient.
  • Random forest hyperparameters = Tuned by FLAML with 10 minute budget, min 500 trees
    Tuned on the full sample and used for all nuisance estimates; affects the DML scores and resulting effects.
  • Propensity trimming rule = Removes 7% of observations, final N = 177,856
    Extension of minmax trimming to the sequential setting; changes the estimand to the trimmed subpopulation.
  • Number of cross-fitting folds K = 5
    Chosen by the researcher following standard DML practice; affects variance and finite-sample properties.
assumptions (5)
  • domain assumption SUTVA (Assumption 2.1): no interference, observed outcomes equal potential outcomes under the assigned sequence.
    Standard in program evaluation, invoked at the start of Section 2.1 to link observed and potential variables.
  • domain assumption Sequential conditional independence (Assumption 2.5[a]): D1 independent of (Y^{d2}, V^{d1}_1) given X0; D2 independent of Y^{d2} given X1, D1 = g1(V0).
    Unverifiable core identification assumption for dynamic policies under dynamic confounding, discussed in Sections 2.5 and 4.5.
  • domain assumption Overlap (Assumption 2.5[b]): positive propensity scores for g1 and g2 given covariate history.
    Required for inverse propensity reweighting; the paper enforces an approximate version by trimming 7 percent of observations in Section 4.6.
  • ad hoc to paper Within-period dynamics are disregarded; individuals in multiple programs within a period are assigned to the longest program.
    Simplification stated in Section 4.4; if within-period assignment is dynamic, the defined treatment states may misrepresent actual sequences.
  • domain assumption Exogeneity of intermediate decision variables: no unmeasured confounders affect D1 and the potential intermediate employment exit Y^{d1}_1.
    Required because the dynamic policy depends on the potential intermediate outcome, as discussed after Assumption 2.5 and in Section 4.5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluating Program Sequences with Double Machine Learning: An Application to Labor Market Policies." pith.science (2026). https://pith.science/paper/5ADREDEQ

@misc{pith2026250611960,
  author       = {Pith},
  title        = {Pith review of: Evaluating Program Sequences with Double Machine Learning: An Application to Labor Market Policies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5ADREDEQ}},
  note         = {Machine review of arXiv:2506.11960}
}
read the original abstract

Many programs evaluated in observational studies incorporate a sequential structure, where individuals may be assigned to various programs over time. While this complexity is often simplified by analyzing programs at single points in time, this paper reviews, explains, and applies methods for program evaluation within a sequential framework. It outlines the assumptions required for identification under dynamic confounding and demonstrates how extending sequential estimands to dynamic policies enables the construction of more realistic counterfactuals. Furthermore, the paper explores recently developed methods for estimating effects across multiple treatments and time periods, utilizing Double Machine Learning (DML), a flexible estimator that avoids parametric assumptions while preserving desirable statistical properties. Using Swiss administrative data, the methods are demonstrated through an empirical application assessing the participation of unemployed individuals in active labor market policies, where assignment decisions by caseworkers can be reconsidered between two periods. The analysis identifies a temporary wage subsidy as the most effective intervention, on average, even after adjusting for its extended duration compared to other programs. Overall, DML-based analysis of dynamic policies proves to be a useful approach within the program evaluation toolkit.

Figures

Figures reproduced from arXiv: 2506.11960 by the authors.

Figure 1
Figure 1. Causal pathways in the sequential treatment effect model with two time periods D1 X1 D2 X0 Y Notes: The arrows in the diagram illustrate the allowed causal pathways be￾tween the variables in the sequential model with dynamic confounding. At any point in time t, covariates Xt and treatments Dt may influence any future treat￾ment, covariate, or outcome. If one of the blue bold arrows is deleted from the figure, confou… view at source ↗
Figure 2
Figure 2. illustrates the distribution of treatment states across the two periods, revealing insights about program size and duration. In the first period, temporary wage subsidies comprise nearly half of the beneficiaries, followed by job-search assistance. Most recipients of temporary wage subsidies in the first period continue in the second period, while recipients of job-search assistance often transition to other program… view at source ↗
Figure 3
Figure 3. GATE-ATE by local language knowledge and previous program participation Fluent Good Intermediate None to basic -3 -2 -1 0 1 2 WS EP TC JA UE no program Not UE -3 -2 -1 0 1 2 -3 -2 -1 0 1 2 -3 -2 -1 0 1 2 -3 -2 -1 0 1 2 -3 -2 -1 0 1 2 GATE-ATE JA-JA vs. TC-TC JA-JA vs. EP-EP JA-JA vs. WS-WS TC-TC vs. EP-EP TC-TC vs. WS-WS EP-EP vs. WS-WS Local language knowledge Unemployed or program in last 5 years Notes: This figur… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 68 canonical work pages

  1. [1]

    \ Cattaneo, M D

    abadie2018econometric APACrefauthors Abadie, A. \ Cattaneo, M D. APACrefauthors \ 2018 . Econometric methods for program evaluation Econometric methods for program evaluation . Annual Review of Economics 10 1 465--503

    work page 2018

  2. [2]

    \ Wager, S

    Athey:2021 APACrefauthors Athey, S. \ Wager, S. APACrefauthors \ 2021 . Policy learning with observational data Policy learning with observational data . Econometrica 89 1 133--161

    work page 2021

  3. [3]

    , Schacht, O

    bach2024hyperparameter APACrefauthors Bach, P. , Schacht, O. , Chernozhukov, V. , Klaassen, S. \ Spindler, M. APACrefauthors \ 2024 . Hyperparameter Tuning for Causal Inference with Double Machine Learning: A Simulation Study Hyperparameter tuning for causal inference with double machine learning: A simulation study . P roceedings of the T hird C onferenc...

    work page 2024

  4. [4]

    \ Robins, J

    bang2005doubly APACrefauthors Bang, H. \ Robins, J. APACrefauthors \ 2005 . Doubly robust estimation in missing data and causal inference models Doubly robust estimation in missing data and causal inference models . Biometrics 61 4 962--973

    work page 2005

  5. [5]

    , Huber, M

    bodory2022evaluating APACrefauthors Bodory, H. , Huber, M. \ Laff \'e rs, L. APACrefauthors \ 2022 . Evaluating (weighted) dynamic treatment effects by double machine learning Evaluating (weighted) dynamic treatment effects by double machine learning . The Econometrics Journal 25 3 628--648

    work page 2022

  6. [6]

    bradic2021high APACrefauthors Bradic, J. , Ji, W. \ Zhang, Y. APACrefauthors \ 2024 . High-dimensional inference for dynamic treatment effects High-dimensional inference for dynamic treatment effects . The Annals of Statistics 52 2 415--440

    work page 2024

  7. [7]

    , Mahlstedt, R

    caliendo:2017 APACrefauthors Caliendo, M. , Mahlstedt, R. \ Mitnik, O A. APACrefauthors \ 2017 . Unobservable, but unimportant? T he relevance of usually unobserved variables for the evaluation of labor market policies Unobservable, but unimportant? T he relevance of usually unobserved variables for the evaluation of labor market policies . Labour Economi...

    work page 2017

  8. [8]

    , Kluve, J

    card2010active APACrefauthors Card, D. , Kluve, J. \ Weber, A. APACrefauthors \ 2010 . Active labour market policy evaluations: A meta-analysis Active labour market policy evaluations: A meta-analysis . The Economic Journal 120 548 F452--F477

    work page 2010

Show all 71 references
  1. [9]

    , Kluve, J

    card2018works APACrefauthors Card, D. , Kluve, J. \ Weber, A. APACrefauthors \ 2018 . What works? A meta analysis of recent active labor market program evaluations What works? A meta analysis of recent active labor market program evaluations . Journal of the European Economic ...

    2018

  2. [10]

    , Chetverikov, D

    chernozhukov2018double APACrefauthors Chernozhukov, V. , Chetverikov, D. , Demirer, M. , Duflo, E. , Hansen, C. , Newey, W. \ Robins, J. APACrefauthors \ 2018 . Double/debiased machine learning for treatment and structural parameters Double/debiased machine learning for treatm...

    2018

  3. [11]

    , Hansen, C

    chernozhukov2024applied APACrefauthors Chernozhukov, V. , Hansen, C. , Kallus, N. , Spindler, M. \ Syrgkanis, V. APACrefauthors \ 2024 . Applied causal inference powered by ML and AI Applied causal inference powered by ML and AI . arXiv preprint arXiv:2403.02467

    2024 arXiv

  4. [12]

    , Newey, W

    chernozhukov2022automaticdyn APACrefauthors Chernozhukov, V. , Newey, W. , Singh, R. \ Syrgkanis, V. APACrefauthors \ 2022 . Automatic debiased machine learning for dynamic treatment effects and general nested functionals Automatic debiased machine learning for dynamic treatme...

    2022 arXiv

  5. [13]

    , Newey, W K

    chernozhukov2022automatic APACrefauthors Chernozhukov, V. , Newey, W K. \ Singh, R. APACrefauthors \ 2022 . Automatic debiased machine learning of causal and structural effects Automatic debiased machine learning of causal and structural effects . Econometrica 90 3 967--1027

    2022

  6. [14]

    , Newey, W K

    chernozhukov2023simple APACrefauthors Chernozhukov, V. , Newey, W K. \ Singh, R. APACrefauthors \ 2023 . A simple and general debiased machine learning theorem with finite-sample guarantees A simple and general debiased machine learning theorem with finite-sample guarantees . ...

    2023

  7. [15]

    , Lechner, M

    Cockx:2020 APACrefauthors Cockx, B. , Lechner, M. \ Bollens, J. APACrefauthors \ 2023 . Priority to unemployed immigrants? A causal machine learning evaluation of training in B elgium Priority to unemployed immigrants? A causal machine learning evaluation of training in B elgi...

    2020

  8. [16]

    , Ferracci, M

    crepon2009active APACrefauthors Cr \'e pon, B. , Ferracci, M. , Jolivet, G. \ Van den Berg, G J. APACrefauthors \ 2009 . Active labor market policy effects in a dynamic setting Active labor market policy effects in a dynamic setting . Journal of the European Economic Associati...

    2009

  9. [17]

    \ Van Den Berg, G J

    crepon2016active APACrefauthors Cr \'e pon, B. \ Van Den Berg, G J. APACrefauthors \ 2016 . Active labor market policies Active labor market policies . Annual Review of Economics 8 521--546

    2016

  10. [18]

    APACrefauthors \ 2015

    dengler2015effectiveness APACrefauthors Dengler, K. APACrefauthors \ 2015 . Effectiveness of sequences of One-Euro-Jobs for welfare recipients in G ermany Effectiveness of sequences of one-euro-jobs for welfare recipients in G ermany . Applied Economics 47 57 6170--6190

    2015

  11. [19]

    APACrefauthors \ 2019

    dengler2019effectiveness APACrefauthors Dengler, K. APACrefauthors \ 2019 . Effectiveness of sequences of classroom training for welfare recipients: what works best in W est G ermany? Effectiveness of sequences of classroom training for welfare recipients: what works best in W...

    2019

  12. [20]

    , Brumback, B

    hernan2002estimating APACrefauthors Hern \'a n, M. , Brumback, B. \ Robins, J. APACrefauthors \ 2002 . Estimating the causal effect of zidovudine on CD4 count with a marginal structural model for repeated measures Estimating the causal effect of zidovudine on cd4 count with a ...

    2002

  13. [21]

    \ Robins, J

    hernan2020causal APACrefauthors Hern \'a n, M. \ Robins, J. APACrefauthors \ 2020 . Causal Inference: What if. Causal inference: What if. Boca Raton Chapman and Hall/CRC

    2020

  14. [22]

    \ Thompson, D J

    horvitz1952generalization APACrefauthors Horvitz, D G. \ Thompson, D J. APACrefauthors \ 1952 . A generalization of sampling without replacement from a finite universe A generalization of sampling without replacement from a finite universe . Journal of the American Statistical...

    1952

  15. [23]

    APACrefauthors \ 2000

    Imbens:2000 APACrefauthors Imbens, G W. APACrefauthors \ 2000 . The role of the propensity score in estimating dose-response functions The role of the propensity score in estimating dose-response functions . Biometrika 87 3 706--710

    2000

  16. [24]

    \ Rubin, D B

    imbens2015causal APACrefauthors Imbens, G W. \ Rubin, D B. APACrefauthors \ 2015 . Causal inference in statistics, social, and biomedical sciences Causal inference in statistics, social, and biomedical sciences . Cambridge University Press

    2015

  17. [25]

    \ Van der Klaauw, B

    kastoryano2022dynamic APACrefauthors Kastoryano, S. \ Van der Klaauw, B. APACrefauthors \ 2022 . Dynamic evaluation of job search assistance Dynamic evaluation of job search assistance . Journal of Applied Econometrics 37 2 227--241

    2022

  18. [26]

    APACrefauthors \ 2024

    kennedy2022semiparametric APACrefauthors Kennedy, E H. APACrefauthors \ 2024 . Semiparametric doubly robust targeted double machine learning: a review Semiparametric doubly robust targeted double machine learning: a review . Handbook of Statistical Methods for Precision Medici...

    2024

  19. [27]

    \ Tamer, E

    khan2010irregular APACrefauthors Khan, S. \ Tamer, E. APACrefauthors \ 2010 . Irregular identification, support conditions, and inverse weight estimation Irregular identification, support conditions, and inverse weight estimation . Econometrica 78 6 2021--2042

    2010

  20. [28]

    APACrefauthors \ 2022

    knaus2022double APACrefauthors Knaus, M C. APACrefauthors \ 2022 . Double machine learning-based programme evaluation under unconfoundedness Double machine learning-based programme evaluation under unconfoundedness . The Econometrics Journal 25 3 602--627

    2022

  21. [29]

    , Lechner, M

    knaus2021machine APACrefauthors Knaus, M C. , Lechner, M. \ Strittmatter, A. APACrefauthors \ 2021 . Machine learning estimation of heterogeneous causal effects: Empirical monte carlo evidence Machine learning estimation of heterogeneous causal effects: Empirical monte carlo e...

    2021

  22. [30]

    \ S tef \'a nik, M

    laffers2024locking APACrefauthors Laff \'e rs, L. \ S tef \'a nik, M. APACrefauthors \ 2024 . Locking in Or Pushing Out: the Caseworker Dilemma Locking in or pushing out: the caseworker dilemma . Institute of Economic Research SAS

    2024

  23. [31]

    APACrefauthors \ 2001

    lechner2001identification APACrefauthors Lechner, M. APACrefauthors \ 2001 . Identification and estimation of causal effects of multiple treatments under the conditional independence assumption Identification and estimation of causal effects of multiple treatments under the co...

    2001

  24. [32]

    APACrefauthors \ 2009

    lechner2009sequential APACrefauthors Lechner, M. APACrefauthors \ 2009 . Sequential causal models for the evaluation of labor market programs Sequential causal models for the evaluation of labor market programs . Journal of Business & Economic Statistics 27 1 71--83

    2009

  25. [33]

    \ Mareckova, J

    lechner2024comprehensive APACrefauthors Lechner, M. \ Mareckova, J. APACrefauthors \ 2024 . Comprehensive Causal Machine Learning Comprehensive causal machine learning . arXiv preprint arXiv:2405.10198

    2024 arXiv

  26. [34]

    \ Miquel, R

    lechner2001potential APACrefauthors Lechner, M. \ Miquel, R. APACrefauthors \ 2001 . A potential outcome approach to dynamic programme evaluation: Nonparametric identification A potential outcome approach to dynamic programme evaluation: Nonparametric identification . Universi...

    2001

  27. [35]

    \ Miquel, R

    lechner2010identification APACrefauthors Lechner, M. \ Miquel, R. APACrefauthors \ 2010 . Identification of the effects of dynamic treatments by sequential conditional independence assumptions Identification of the effects of dynamic treatments by sequential conditional indepe...

    2010

  28. [36]

    \ Strittmatter, A

    lechner2019practical APACrefauthors Lechner, M. \ Strittmatter, A. APACrefauthors \ 2019 . Practical procedures to deal with common support problems in matching estimation Practical procedures to deal with common support problems in matching estimation . Econometric Reviews 38...

    2019

  29. [37]

    \ Wiehler, S

    lechner2013does APACrefauthors Lechner, M. \ Wiehler, S. APACrefauthors \ 2013 . Does the order and timing of active labour market programmes matter? Does the order and timing of active labour market programmes matter? Oxford Bulletin of Economics and Statistics 75 2 180--212

    2013

  30. [38]

    \ Wunsch, C

    lechner2013sensitivity APACrefauthors Lechner, M. \ Wunsch, C. APACrefauthors \ 2013 . Sensitivity of matching-based program evaluations to the availability of control variables Sensitivity of matching-based program evaluations to the availability of control variables . Labour...

    2013

  31. [39]

    \ Syrgkanis, V

    lewis2020double APACrefauthors Lewis, G. \ Syrgkanis, V. APACrefauthors \ 2021 . Double/Debiased Machine Learning for Dynamic Treatment Effects via g-estimation Double/debiased machine learning for dynamic treatment effects via g-estimation . Advances in Neural Information Pro...

    2021

  32. [40]

    , Bearth, N

    mascolo2024heterogeneous APACrefauthors Mascolo, F. , Bearth, N. , Muny, F. , Lechner, M. \ Mareckova, J. APACrefauthors \ 2024 . From Average Effects to Targeted Assignment: A Causal Machine Learning Analysis of Swiss Active Labor Market Policies From average effects to targe...

    2024 arXiv

  33. [41]

    \ Singh, R

    singh2021finite APACrefauthors Meza, I. \ Singh, R. APACrefauthors \ 2021 . Nested Nonparametric Instrumental Variable Regression: Long Term, Mediated, and Time Varying Treatment Effects Nested nonparametric instrumental variable regression: Long term, mediated, and time varyi...

    2021 arXiv

  34. [42]

    APACrefauthors \ 2003

    murphy2003optimal APACrefauthors Murphy, S A. APACrefauthors \ 2003 . Optimal dynamic treatment regimes Optimal dynamic treatment regimes . Journal of the Royal Statistical Society: Series B (Statistical Methodology) 65 2 331--355

    2003

  35. [43]

    , Brunskill, E

    nie2021learning APACrefauthors Nie, X. , Brunskill, E. \ Wager, S. APACrefauthors \ 2021 . Learning when-to-treat policies Learning when-to-treat policies . Journal of the American Statistical Association 116 533 392--409

    2021

  36. [44]

    APACrefauthors \ 1995

    pearl1995causal APACrefauthors Pearl, J. APACrefauthors \ 1995 . Causal diagrams for empirical research Causal diagrams for empirical research . Biometrika 82 4 669--688

    1995

  37. [45]

    , Varoquaux, G

    pedregosa2011scikit APACrefauthors Pedregosa, F. , Varoquaux, G. , Gramfort, A. , Michel, V. , Thirion, B. , Grisel, O. others APACrefauthors \ 2011 . Scikit-learn: Machine learning in Python Scikit-learn: Machine learning in python . The Journal of Machine Learning Research 1...

    2011

  38. [46]

    \ Rotnitzky, A

    richardson2014causal APACrefauthors Richardson, T S. \ Rotnitzky, A. APACrefauthors \ 2014 . Causal etiology of the research of J ames M . R obins Causal etiology of the research of J ames M . R obins . Statistical Science 29 4 459--484

    2014

  39. [47]

    APACrefauthors \ 1986

    robins1986new APACrefauthors Robins, J. APACrefauthors \ 1986 . A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect A new approach to causal inference in mortality studies with a ...

    1986

  40. [48]

    A new approach to causal inference in mortality studies with sustained exposure periods—application to control of the healthy worker survivor effect

    robins1987addendum APACrefauthors Robins, J. APACrefauthors \ 1987 . Addendum to “A new approach to causal inference in mortality studies with sustained exposure periods—application to control of the healthy worker survivor effect” Addendum to “a new approach to causal inferen...

    1987

  41. [49]

    APACrefauthors \ 1989

    robins1989analysis APACrefauthors Robins, J. APACrefauthors \ 1989 . The analysis of randomized and non-randomized AIDS treatment trials using a new approach to causal inference in longitudinal studies The analysis of randomized and non-randomized aids treatment trials using a...

    1989

  42. [50]

    APACrefauthors \ 1994

    robins1994correcting APACrefauthors Robins, J. APACrefauthors \ 1994 . Correcting for non-compliance in randomized trials using structural nested mean models Correcting for non-compliance in randomized trials using structural nested mean models . Communications in Statistics -...

    1994

  43. [51]

    APACrefauthors \ 2000

    robins2000robust APACrefauthors Robins, J. APACrefauthors \ 2000 . Robust estimation in sequentially ignorable missing data and causal inference models Robust estimation in sequentially ignorable missing data and causal inference models . Proceedings of the A merican S tatisti...

    2000

  44. [52]

    \ Hern \'a n, M

    robins2009estimation APACrefauthors Robins, J. \ Hern \'a n, M. APACrefauthors \ 2008 . Estimation of the causal effects of time-varying exposures Estimation of the causal effects of time-varying exposures . G. Fitzmaurice, M. Davidian, G. Verbeke \ G. Molenberghs\ ( ), Longit...

    2008

  45. [53]

    , Hernan, M A

    robins2000marginal APACrefauthors Robins, J. , Hernan, M A. \ Brumback, B. APACrefauthors \ 2000 . Marginal structural models and causal inference in epidemiology Marginal structural models and causal inference in epidemiology . Epidemiology 11 5 550--560

    2000

  46. [54]

    , Rotnitzky, A

    robins1994estimation APACrefauthors Robins, J. , Rotnitzky, A. \ Zhao, L P. APACrefauthors \ 1994 . Estimation of regression coefficients when some regressors are not always observed Estimation of regression coefficients when some regressors are not always observed . Journal o...

    1994

  47. [55]

    \ Rubin, D B

    rosenbaum1983central APACrefauthors Rosenbaum, P R. \ Rubin, D B. APACrefauthors \ 1983 . The central role of the propensity score in observational studies for causal effects The central role of the propensity score in observational studies for causal effects . Biometrika 70 1 41--55

    1983

  48. [56]

    APACrefauthors \ 1974

    rubin1974estimating APACrefauthors Rubin, D B. APACrefauthors \ 1974 . Estimating causal effects of treatments in randomized and nonrandomized studies. Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Educational Psychology 66 5 688

    1974

  49. [57]

    APACrefauthors \ 2024

    sakaguchi2024robust APACrefauthors Sakaguchi, S. APACrefauthors \ 2024 . Robust Learning for Optimal Dynamic Treatment Regimes with Observational Data Robust learning for optimal dynamic treatment regimes with observational data . arXiv preprint arXiv:2404.00221

    2024 arXiv

  50. [58]

    , Rotnitzky, A

    scharfstein1999adjusting APACrefauthors Scharfstein, D O. , Rotnitzky, A. \ Robins, J. APACrefauthors \ 1999 . Adjusting for nonignorable drop-out using semiparametric nonresponse models Adjusting for nonignorable drop-out using semiparametric nonresponse models . Journal of t...

    1999

  51. [59]

    APACrefauthors \ 2004

    sianesi2004evaluation APACrefauthors Sianesi, B. APACrefauthors \ 2004 . An evaluation of the S wedish system of active labor market programs in the 1990s An evaluation of the S wedish system of active labor market programs in the 1990s . Review of Economics and Statistics 86 ...

    2004

  52. [60]

    , Robins, J M

    taubman2009intervening APACrefauthors Taubman, S L. , Robins, J M. , Mittleman, M A. \ Hern \'a n, M A. APACrefauthors \ 2009 . Intervening on risk factors for coronary heart disease: an application of the parametric g-formula Intervening on risk factors for coronary heart dis...

    2009

  53. [61]

    , Yiannoutsos, C

    tran2019double APACrefauthors Tran, L. , Yiannoutsos, C. , Wools-Kaloustian, K. , Siika, A. , Van Der Laan, M. \ Petersen, M. APACrefauthors \ 2019 . Double robust efficient estimators of longitudinal treatment effects: comparative performance in simulations and a case study D...

    2019

  54. [62]

    \ Vikstr \"o m, J

    van2019long APACrefauthors Van den Berg, G J. \ Vikstr \"o m, J. APACrefauthors \ 2022 . Long-Run Effects of Dynamically Assigned Treatments: A New Methodology and an Evaluation of Training Effects on Earnings Long-run effects of dynamically assigned treatments: A new methodol...

    2022

  55. [63]

    \ Gruber, S

    van2012targeted APACrefauthors Van der Laan, M J. \ Gruber, S. APACrefauthors \ 2012 . Targeted minimum loss based estimation of causal effects of multiple time point interventions Targeted minimum loss based estimation of causal effects of multiple time point interventions . ...

    2012

  56. [64]

    \ Rubin, D

    van2006targeted APACrefauthors Van Der Laan, M J. \ Rubin, D. APACrefauthors \ 2006 . Targeted maximum likelihood learning Targeted maximum likelihood learning . The International Journal of Biostatistics 2 1

    2006

  57. [65]

    APACrefauthors \ 2017

    vikstrom2017dynamic APACrefauthors Vikstr \"o m, J. APACrefauthors \ 2017 . Dynamic treatment assignment and evaluation of active labor market policies Dynamic treatment assignment and evaluation of active labor market policies . Labour Economics 49 42--54

    2017

  58. [66]

    , Haelermans, C

    vooren2019effectiveness APACrefauthors Vooren, M. , Haelermans, C. , Groot, W. \ Maassen van den Brink, H. APACrefauthors \ 2019 . The effectiveness of active labor market policies: a meta-analysis The effectiveness of active labor market policies: a meta-analysis . Journal of...

    2019

  59. [67]

    APACrefauthors \ 2024

    wager2024causal APACrefauthors Wager, S. APACrefauthors \ 2024 . Causal Inference: A Statistical Learning Approach Causal inference: A statistical learning approach . Stanford University

    2024

  60. [68]

    wang2021flaml APACrefauthors Wang, C. , Wu, Q. , Weimer, M. \ Zhu, E. APACrefauthors \ 2021 . FLAML : A fast and lightweight automl library FLAML : A fast and lightweight automl library . Proceedings of Machine Learning and Systems 3 434--447

    2021

  61. [69]

    \ Dalton, J E

    yang2012unified APACrefauthors Yang, D. \ Dalton, J E. APACrefauthors \ 2012 . A unified approach to measuring the effect size between two groups using SAS A unified approach to measuring the effect size between two groups using SAS . SAS global forum SAS global forum \ ( 335, \ 1--6)

    2012

  62. [70]

    , Cain, L E

    young2011comparative APACrefauthors Young, J G. , Cain, L E. , Robins, J M. , O’Reilly, E J. \ Hern \'a n, M A. APACrefauthors \ 2011 . Comparative effectiveness of dynamic treatment regimes: an application of the parametric g-formula Comparative effectiveness of dynamic treat...

    2011

  63. [71]

    , Tsiatis, A A

    zhang2013robust APACrefauthors Zhang, B. , Tsiatis, A A. , Laber, E B. \ Davidian, M. APACrefauthors \ 2013 . Robust estimation of optimal dynamic treatment regimes for sequential treatment decisions Robust estimation of optimal dynamic treatment regimes for sequential treatme...

    2013

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.