Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Semiparametric Inference for Partially Identifiable Data Fusion Estimands via Double Machine Learning

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read For decomposable data-fusion estimands, the paper proves that two conditional moments define valid outer bounds on the partial-identification region and that cross-fitted double-machine-learning estimators of the endpoints are…

desk verdict Solid, useful data-fusion method with a real proof gap in Theorem 2's appendix; likely fixable, but the paper as posted doesn't establish the headline coverage guarantee. read the letter →

arxiv 2502.05319 v1 pith:KGYZNOSB submitted 2025-02-07 stat.ME

Yicong Jiang , Lucas Janson This is my paper · ORCID
classification stat.ME MSC 62D1062G0562G2062F12
keywords datafusionpartialidentificationdoublemachinelearningsemiparametricefficiencyCauchy-Schwarzboundsconditionalmomentsconfidenceintervalsmissingatrandom
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

This paper addresses a common data-fusion situation: two independent samples observe $(Y,X)$ and $(Z,X)$ but never $(Y,Z,X)$ together, so a target $\theta=E[h(Y,Z,X)]$ is only partially identifiable. For the large class of decomposable targets $h(y,z,x)=f(y,x)^T g(z,x)$, the authors prove that the entire partial-identification region is trapped inside an outer interval whose endpoints are explicit functions of the conditional means and variances of $f(Y,X)$ and $g(Z,X)$. They then build cross-fitted double-machine-learning estimators of those endpoints, prove asymptotic normality and semiparametric efficiency, and give consistent variance estimators, so the reported interval covers $\theta$ with asymptotic probability at least $1-\alpha$. If correct, an analyst needs only standard conditional-mean regression software to obtain valid, near-sharp confidence statements about quantities no single dataset can reveal.

What carries the argument

The load-bearing object is the Cauchy-Schwarz bound: after standardizing $f(Y,X)$ and $g(Z,X)$ by their conditional means and variances, the cross-term $E[f^T g]$ is bounded, conditional on $X$, by the product of the two conditional standard-deviation matrices, and Theorem 1 shows this bound is tight within the class of distributions that match only the first two conditional moments. The inference machinery is the efficient influence function of the bound endpoints, which contains terms like $(f-m_Y)m_Z$ and $((f-m_Y)^2-v_Y)\sqrt{v_Z/v_Y}$; adding its plug-in estimate to the cross-fitted plug-in estimator yields Neyman orthogonality, so nuisance estimation error enters only through product terms, and the estimator attains the semiparametric variance bound.

What would settle it

Run Algorithm 1 on data where one dataset has an $X$-stratum with exactly zero conditional variance, for instance $Y=\beta^T X$ with no noise while $Z$ has noise; the estimated interval's width will blow up and the empirical coverage of the nominal 95% interval for $\theta$ will fall below 0.95 even at large sample sizes, directly contradicting the coverage corollary of Theorems 2 and 3 under Assumption 3.

Watch

Extended reading notes

Core claim

The central discovery is Theorem 1: if $h$ is $(f,g)$-decomposable, then the second-order (Cauchy-Schwarz) bounds have closed forms $\theta^{(\mathrm{CS})}_L = E[m_Y(X)^T m_Z(X)] - E[\operatorname{tr}(\sqrt{\operatorname{Var}(g\mid X)}\,\operatorname{Var}(f\mid X)\,\sqrt{\operatorname{Var}(g\mid X)})]$ and symmetrically for the upper bound, where $m_Y,m_Z$ are conditional means and $\operatorname{Var}(f\mid X),\operatorname{Var}(g\mid X)$ are conditional covariance matrices. These bounds are outer bounds on the tight partially identifiable region $[\theta_L,\theta_U]$, and the paper shows they coincide with the tight region whenever the conditional distributions of $f(Y,X)\mid X$ and $g(Z,X)\mid X$ agree up to $X$-dependent location and scale (Proposition 1). Theorems 2 and 3 show that the cross-fitted double-machine-learning estimators of the two endpoints are $\sqrt{n}$-consistent, asymptotically normal, and semiparametrically efficient with consistently estimable variances, which yields an asymptotically valid $1-\alpha$ confidence interval for the original estimand $\theta$.

Load-bearing premise

The confidence-interval guarantee depends on the conditional variances of $f(Y,X)\mid X$ and $g(Z,X)\mid X$ being safely bounded away from zero, with finite inverse-moment conditions, so that terms like $1/\sqrt{\text{conditional variance}}$ in the influence function do not explode; if some $X$-stratum is nearly deterministic in either dataset, the asymptotic normality proof breaks down and coverage is not guaranteed.

Editorial extensions

If this is right

  • For any decomposable estimand, valid inference is possible with only conditional mean and variance regressions, so standard supervised-learning tools suffice.
  • The method handles continuous and multivariate $Y$ and $Z$, going beyond earlier binary or one-dimensional data-fusion inference.
  • When the two conditional distributions are location-scale versions of each other, the reported interval is as narrow as the tight partial-identification region.
  • Estimands including $E[YZ]$, the average relative treatment effect $E[Y/Z]$, the variance of individual treatment effects, subgroup effects, and Lee bounds all fit the decomposable framework.
  • The confidence interval covers the original partially identified estimand $\theta$ with asymptotic probability at least $1-\alpha$, not just the outer-bound endpoints.

Reading between the lines

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

  • A natural extension is to treat the outer bounds as a sensitivity analysis for unconfoundedness: if the missing-at-random assumption fails, the reported interval is no longer guaranteed to contain $\theta$, and one could quantify how violations shift the bounds.
  • The tightness condition (location-scale agreement) could be checked empirically by comparing conditional quantile functions or using goodness-of-fit tests, giving practitioners a diagnostic for when the interval is sharp versus merely valid.
  • Because only first and second conditional moments enter, the method should combine naturally with high-dimensional sparse regression or deep learning for the nuisance functions, provided the required $o(n^{-1/4})$ rates hold; the heavy 16th-moment assumptions suggest caution with very heavy-tailed outcomes.
  • The delta-method extension for ratios indicates the framework extends beyond single expectations to smooth functions of several identifiable and partially identifiable components, such as the OLS coefficient in the empirical application.
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

3 major / 5 minor

Summary. The paper studies data fusion settings where (Y,X) and (Z,X) are observed in separate samples and the target is a partially identifiable estimand θ = E[h(Y,Z,X)]. It proposes Cauchy-Schwarz outer bounds for the partial-identification region that depend only on conditional first and second moments of f(Y,X)|X and g(Z,X)|X, proves conditions under which the bounds are tight, and constructs cross-fitted DML estimators of the two endpoints with plug-in variance estimators. The main theoretical claims are asymptotic normality and semiparametric efficiency of these endpoint estimators (Theorems 2 and 3), leading to a confidence interval for θ. The method is compared with Ji et al. (2023) in simulations and applied to a consumption-wealth data fusion problem.

Significance. If the theorems are established, this is a useful and practical contribution: the bounds have closed forms, the implementation requires only standard conditional-mean regression, the DML framework is appropriate for machine-learning nuisance estimates, and the empirical comparisons with Dualbounds are informative. The authors also provide reproducible code and openly discuss limitations such as variance positivity. However, the proof of the central asymptotic-normality theorem currently contains a concrete algebraic error at the step that establishes the required bias rate, so the coverage guarantee of Algorithm 1 is not supported by the manuscript as written. The result appears repairable, but the proof must be corrected before the claims can be accepted.

major comments (3)
  1. [Appendix I, Eqs. (63) and (66)] The proof of Theorem 2 needs the bias term T_{2k} to be o_p(n^{-1/2}), and the bound on U_{24k} in Eq. (66) is the load-bearing step. As printed, that equality is algebraically incorrect. Writing a = \hat v_Y/v_Y and b = \hat v_Z/v_Z, definition (63) gives U_{24k} = E[\sqrt{v_Y v_Z}(\sqrt{a/b}+\sqrt{b/a}-2)] = E[\sqrt{v_Y v_Z}(\sqrt a-\sqrt b)^2 / \sqrt{ab}], while the right-hand side of Eq. (66) equals E[\sqrt{\hat v_Y \hat v_Z}(\sqrt{\hat v_Y/v_Y}-\sqrt{\hat v_Z/v_Z})^2] = E[\sqrt{v_Y v_Z}\sqrt{ab}(\sqrt a-\sqrt b)^2], which differs by a factor of ab; the factor 1/2 in the first line of (66) is also not present in (63). The correct identity is U_{24k} = E[\sqrt{\hat v_Y \hat v_Z}(\sqrt{v_Y/\hat v_Y}-\sqrt{v_Z/\hat v_Z})^2]. Because the rate U_{24k}=o(n^{-1/2}) is essential for Theorem 2 and hence for the coverage guarantee of Algorithm 1, this proof step must be rewritten. I expect the rate claim is true under Assumptions 3 and 5 via a second-order Taylor expansion, but the proof as printed does not establish it.
  2. [Section 2.2, Theorem 1, Eqs. (5)-(6)] The matrix expression in the displayed Cauchy-Schwarz bound needs correction. As printed, the second term appears to be E[tr(\sqrt{\mathrm{Var}[g]\mathrm{Var}[f]}\sqrt{\mathrm{Var}[g]})], which in the scalar case equals v_Z\sqrt{v_Y} rather than the required \sqrt{v_Y v_Z}. The proof in Appendix G derives the trace of the singular values of \sqrt{\mathrm{Var}[f]}\sqrt{\mathrm{Var}[g]}, i.e., tr(\sqrt{\sqrt{\mathrm{Var}[g]}\,\mathrm{Var}[f]\sqrt{\mathrm{Var}[g]}}), with the outer square root applied to the whole matrix. The displayed theorem should be made consistent with that derivation; as it stands, the object being estimated is not defined correctly.
  3. [Theorem 2 and Appendix I] The statement that the estimators are semiparametrically efficient is asserted but not proved: the proof establishes asymptotic normality with variance equal to the variance of the claimed influence function, but it does not identify the semiparametric tangent space or show that this variance attains the semiparametric lower bound. This does not affect the validity of the confidence interval, but it is part of the paper's central claim, so either a tangent-space argument should be supplied or the statement should be softened to asymptotic normality with the stated influence-function variance.
minor comments (5)
  1. [Section 3.2] There is a typo in the sentence introducing Algorithm 1: 'Algirhtm' should be 'Algorithm 1'.
  2. [Assumption 3 and Section 3.2] The conditional variance positivity assumption, E[1/Var(f|X)^8] < ∞ and E[1/Var(g|X)^8] < ∞, is an untestable restriction that fails when f or g is nearly deterministic on any stratum. Appendix D gives a helpful example, but the main text should state more prominently that confidence-interval validity can break down in such cases.
  3. [Section 4.1.1 and Figure 1] The x-axis label in Figure 1 is printed as 'σY σZ' but the text describes varying the ratio σY/σZ; please check that the label, axis, and caption are consistent.
  4. [References] Several references contain typos, including 'Santos-Lozadab', 'Techincal Report', and the encoding of 'Rässler'; these should be corrected in the final version.
  5. [Section 3.1 and Algorithm 1] The notation for the cross-fitted estimates is verbose and sometimes inconsistent (e.g., \vY, \vZ, and \vz are all used for the same object); a single notational convention would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the bounds and estimators are derived from first principles against identified marginals, with all nuisance parameters estimated from external data.

full rationale

The paper's central derivation chain is not circular. Theorem 1 (Eqs. 5-6) derives closed-form Cauchy-Schwarz bounds on the partially identified estimand from the definition of the second-order relaxation C^(2)_{f,g,x} in Eqs. (3)-(4), using Cauchy-Schwarz on standardized conditional residuals; the proof in Appendix G constructs a joint distribution matching the first two conditional moments to show tightness of the relaxation, so the bound is a theorem rather than an assumed input. Proposition 1 gives a separate, non-circular condition for tightness relative to the sharp bounds. The DML estimators in Algorithm 1 estimate nuisance functions (conditional means, variances, propensity) from held-out folds and then form plug-in and debiased estimators; the target endpoints are not fitted parameters and the simulations use data generated independently of the method's constants. Theorems 2-3 are standard semiparametric efficiency claims justified by influence-function calculations and rate conditions, and the variance estimators are evaluated on the same influence functions. There are no load-bearing self-citations: comparisons with Ji et al. (2023) are contrastive, not justificatory. The positivity assumption (Assumption 3) is a regularity condition limiting applicability, not a circularity. A skeptical note about a possible algebraic error in the U24k step of the proof of Theorem 2 (Appendix I, around Eq. 66) concerns the validity of a proof step, not circular reasoning: even if the displayed equality in Eq. (66) is incorrect, that would be a correctness gap, not a reduction of the conclusion to the assumptions by construction. Therefore the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claim rests on unconfoundedness, positivity of conditional variances, finite higher moments, decomposability of the estimand, and standard semiparametric theory. No free parameters are fitted in the derivation itself; the method requires the analyst to choose f and g so that h = f^T g, which is a modeling choice rather than a fitted constant.

assumptions (6)
  • domain assumption Missing at random: R is independent of (Y,Z) given X (Assumption 1).
    Needed so that P(Y|X,R=1)=P(Y|X) and similarly for Z; without it the two datasets share no identifiable information about the joint distribution.
  • domain assumption Decomposability of h: h(y,z,x)=f(y,x)^T g(z,x) (Definition 1).
    The closed-form Cauchy-Schwarz bounds and the DML estimator require this factorization; for non-decomposable h, the paper relies on basis approximation without a formal error analysis.
  • domain assumption Positivity: E[1/e(X)^4] < infinity and E[1/Var(f|X)^8], E[1/Var(g|X)^8] < infinity (Assumption 3).
    The square root of conditional variance appears in the efficient influence function, so the denominator can blow up; the proofs require these reciprocal moments to be finite.
  • domain assumption Nuisance estimation rates: estimators achieve o(n^{-1/4}) L_p errors (Assumptions 4-5).
    Double machine learning requires these rates for the empirical process and bias terms to vanish; in practice this ties method validity to the ML toolbox chosen.
  • domain assumption Finite 16th moments of f(Y,X), g(Z,X) and nuisance estimators (Assumption 2).
    The proof bounds high-order products of estimation errors using moment assumptions; the 16th moment arises because the estimand is a fourth-order function of the underlying variables.
  • standard math Cauchy-Schwarz inequality and singular value decomposition trace bounds used in Theorem 1.
    The proof of Theorem 1 (Appendix G) uses these results to bound the cross-term E[epsilon_Y^T Sigma^{1/2} epsilon_Z].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Semiparametric Inference for Partially Identifiable Data Fusion Estimands via Double Machine Learning." pith.science (2026). https://pith.science/paper/KGYZNOSB

@misc{pith2026250205319,
  author       = {Pith},
  title        = {Pith review of: Semiparametric Inference for Partially Identifiable Data Fusion Estimands via Double Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KGYZNOSB}},
  note         = {Machine review of arXiv:2502.05319}
}
read the original abstract

Many statistical estimands of interest (e.g., in regression or causality) are functions of the joint distribution of multiple random variables. But in some applications, data is not available that measures all random variables on each subject, and instead the only possible approach is one of data fusion, where multiple independent data sets, each measuring a subset of the random variables of interest, are combined for inference. In general, since all random variables are never observed jointly, their joint distribution, and hence also the estimand which is a function of it, is only partially identifiable. Unfortunately, the endpoints of the partially identifiable region depend in general on entire conditional distributions, rendering them hard both operationally and statistically to estimate. To address this, we present a novel outer-bound on the region of partial identifiability (and establish conditions under which it is tight) that depends only on certain conditional first and second moments. This allows us to derive semiparametrically efficient estimators of our endpoint outer-bounds that only require the standard machine learning toolbox which learns conditional means. We prove asymptotic normality and semiparametric efficiency of our estimators and provide consistent estimators of their variances, enabling asymptotically valid confidence interval construction for our original partially identifiable estimand. We demonstrate the utility of our method in simulations and a data fusion problem from economics.

Figures

Figures reproduced from arXiv: 2502.05319 by the authors.

Figure 1
Figure 1. Influence of the noise level σY on our method’s and Dualbounds’ coverage (left) and width (right) for 95% confidence intervals in the simulation of Section 4.1.1. Error bars represent ±1.96 Monte Carlo standard errors. f(Y, X) = Y and g(Z, X) = 1/Z and our (unidentifiable) estimand3 is θ = E [Y/Z] = exp{σ 2 (1 − ρ) + 0.5(β1 − β0) TΣ(β1 − β0)}. (8) For this simulation, both our method and Dualbounds are not given the… view at source ↗
Figure 2
Figure 2. Influence of the noise level σ on our method’s and Dualbounds’ coverage (left) and width (right) for 95% confidence intervals in the simulation of Section 4.1.2. Error bars represent ±1.96 Monte Carlo standard errors. 4.2 Relating Consumption to Wealth We consider the data fusion problem introduced in Bostic et al. (2009) as described in Section 1, and which was also analyzed in Evans et al. (2018). Recall that to s… view at source ↗
Figure 3
Figure 3. Influence of the noise level σY on our method’s and Dualbounds’ coverage (left) and width (right) for 95% confidence intervals in the simulation of Appendix D. Error bars represent ±1.96 Monte Carlo standard errors. cancer and the infection history of certain viruses. Ideally, one hopes to directly study the correlation between the gold standard, the cancer precursor conditions Y , and the precise medical record Z. … view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Influence of the noise level σ and the correlation parameter ρ on coverage in the simulation of Appendix E.2. Error bars represent ±1.96 Monte Carlo standard errors. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.5 1.0 1.5 2.0 0.25 0.50 0.75 1.00 σ Width (a) ρ = 0 ● ● ● ● ●…
Figure 5
Figure 5. Figure 5: Influence of the noise level σ and the correlation parameter ρ on width.“Theory” stands for the difference of the upper and lower bounds.“CI” stands for the difference of the upper confidence bound for the upper bound and the lower confidence bound for the lower bound.…
Figure 6
Figure 6. Figure 6: The violin plots of the LCB and UCB with varying [PITH_FULL_IMAGE:figures/full_fig_p031_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Partial identification via conditional linear programs: estimation and policy learning

    stat.ME 2025-06 conditional novelty 7.0 of 10

    Two debiased estimators, one based on linear programming solutions and one on entropic smoothing, provide asymptotic confidence intervals for covariate-dependent partial identification bounds and support policy learning.

Reference graph

Works this paper leans on

80 extracted references · 69 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...

  2. [2]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sentence := #2 '...

  3. [3]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  4. [4]

    Anderberg, M. R. (1983). Cluster Analysis for Applications . New York: Academic Press

    work page 1983

  5. [5]

    and Carroll, J

    Arabie, P. and Carroll, J. D. (1980). Mapclus: A mathematical programming approach to fitting the adclus models. Psychometrika 445, 211--235

    work page 1980

  6. [6]

    Ball, G. H. and Hall, D. J. (1965). A novel method of data analysis and pattern classification. Technical report, Stanford Research Institute, California

    work page 1965

  7. [7]

    Banfield, J. D. and Raftery, A. E. (1993). Model-based gaussian and non-gaussian clustering. Biometrics 49, 803--821

    work page 1993

  8. [8]

    Beale, E. M. L. (1969). Euclidean cluster analysis. Bulletin of the International Statistical Institute 43, 92--94

    work page 1969

Show all 80 references
  1. [9]

    Bezdek, J. C. (1974). Numerical taxonomy with fuzzy sets. Journal of Methematical Biology 1, 57--71

    1974

  2. [10]

    Binder, D. A. (1978). Bayesian cluster analysis. Biometrika 65, 31--38

    1978

  3. [11]

    Blashfield, R. K. (1976). Mixture model tests of cluster analysis: Accuracy of four agglomerative hierarchical methods. Psychological Bulletin 83, 377--385

    1976

  4. [12]

    Breiman, L., Friedman, J., Olshen, R., and Stone, C. (1984). Classification and Regression Trees . Belmont: Wadsworth

    1984

  5. [13]

    and Harabasz, J

    Calinski, T. and Harabasz, J. (1974). A dendrite method for cluster analysis. Communications in Statistics 3, 1--27

    1974

  6. [14]

    G., Fraley, C., Murtagh, F., and Raftery, A

    Campbell, J. G., Fraley, C., Murtagh, F., and Raftery, A. E. (1997). Linear flaw detection in woven textiles using model-based clustering. Pattern Recognition Letters 18, 1539--1548

    1997

  7. [15]

    G., Fraley, C., Stanford, D., Murtagh, F., and Raftery, A

    Campbell, J. G., Fraley, C., Stanford, D., Murtagh, F., and Raftery, A. E. (1999). Model-based methods for real-time textile fault detection. International Journal of Imaging Systems and Technology 10, 339--346

    1999

  8. [16]

    W., Gorge, L

    Carmichael, J. W., Gorge, L. A., and Julius, R. S. (1968). Finding natural clusters. Systematic Zoology 17, 144--150

    1968

  9. [17]

    Carmichael, J. W. and Sneath, P. H. A. (1969). Taxometric maps. Systematic Zoology 18, 402--415

    1969

  10. [18]

    Cattell, R. B. and Coulter, M. A. (1966). Principles of behavioural taxonomy and the mathematical basis of the taxonome computer program. British Journal of Mathematical and Statistical Psychology 19, 237--269

    1966

  11. [19]

    Cormack, R. (1971). A review of classification. Journal of the Royal Statistical Society A 134, 321--367

    1971

  12. [20]

    Crawford, R. M. M. and Wishart, D. (1976). A rapid multivariate method for the detection and classification of groups of ecologically related species. Journal of Economics 55, 505--524

    1976

  13. [21]

    and Raftery, A

    Dasgupta, A. and Raftery, A. E. (1998). Detecting features in spatial point processes with clutter via model-based clustering. Journal of the American Statistical Association 93, 294--302

    1998

  14. [22]

    Day, N. E. (1969). Estimating the components of a mixture of normal distributions. Biometrika 56, 463--474

    1969

  15. [23]

    De Soete, G. (1986). Optimal variable weighting for ultrametric and additive tree clustering. Quality and Quantity 20, 169--180

    1986

  16. [24]

    P., Laird, N

    Dempster, A. P., Laird, N. M., and Rubin, D. B. (1977). Maximum likelihood for incomplete data via the em algorithm. Journal of the Royal Statistical Society B 39, 1--38

    1977

  17. [25]

    Cheng, and D

    Baiocchi, M., J. Cheng, and D. S. Small (2014). Instrumental variable methods for causal inference. Statistics in medicine\/ 33\/ (13), 2297--2340

    2014

  18. [26]

    Molchanov, and F

    Beresteanu, A., I. Molchanov, and F. Molinari (2012). Partial identification using random set theory. Journal of Econometrics\/ 166\/ (1), 17--32

    2012

  19. [27]

    Bickel, P. J., C. A. Klaassen, P. J. Bickel, Y. Ritov, J. Klaassen, J. A. Wellner, and Y. Ritov (1993). Efficient and adaptive estimation for semiparametric models , Volume 4. Springer

    1993

  20. [28]

    Gabriel, and G

    Bostic, R., S. Gabriel, and G. Painter (2009). Housing wealth, financial wealth, and consumption: New evidence from micro data. Regional Science and Urban Economics\/ 39\/ (1), 79--89

    2009

  21. [29]

    Castanedo, F. et al. (2013). A review of data fusion techniques. The scientific world journal\/ 2013

    2013

  22. [30]

    Chen, J. and J. Roth (2023). Logs with zeros? some problems and solutions

    2023

  23. [31]

    Chetverikov, M

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

    2018

  24. [32]

    Cho, W. T. and C. F. Manski (2008a, 08). 547 Cross‐Level/Ecological Inference . In The Oxford Handbook of Political Methodology . Oxford University Press

  25. [33]

    Cho, W. T. and C. F. Manski (2008b). Cross-level/ecological inference

    2008

  26. [34]

    Cross, P. J. and C. F. Manski (2002). Regressions, short and long. Econometrica\/ 70\/ (1), 357--368

    2002

  27. [35]

    D \' az, I. (2017). Efficient estimation of quantiles in missing data models. Journal of Statistical Planning and Inference\/ 190 , 39--51

    2017

  28. [36]

    Di Zio, and M

    D'Orazio, M., M. Di Zio, and M. Scanu (2006). Statistical matching: Theory and practice . John Wiley & Sons

    2006

  29. [37]

    Evans, K., B. Sun, J. Robins, and E. J. T. Tchetgen (2018). Doubly robust regression analysis for data fusion. arXiv preprint arXiv:1808.07309\/

    2018 arXiv

  30. [38]

    potential outcomes

    Fan, Y., E. Guerre, and D. Zhu (2017a). Partial identification of functionals of the joint distribution of “potential outcomes”. Journal of Econometrics\/ 197\/ (1), 42--59

    2017

  31. [39]

    potential outcomes

    Fan, Y., E. Guerre, and D. Zhu (2017b). Partial identification of functionals of the joint distribution of “potential outcomes”. Journal of econometrics\/ 197\/ (1), 42--59

    2017

  32. [40]

    Sherman, and M

    Fan, Y., R. Sherman, and M. Shum (2016). Estimation and inference in an ecological inference model. Journal of Econometric Methods\/ 5\/ (1), 17--48

    2016

  33. [41]

    Firpo, S. and G. Ridder (2019). Partial identification of the treatment effect distribution and its functionals. Journal of Econometrics\/ 213\/ (1), 210--234

    2019

  34. [42]

    Fox, M. P., T. L. Lash, and L. M. Bodnar (2020). Common misconceptions about validation studies. International Journal of Epidemiology\/ 49\/ (4), 1392--1396

    2020

  35. [43]

    Frank, M. J., R. B. Nelsen, and B. Schweizer (1987). Best-possible bounds for the distribution of a sum—a problem of kolmogorov. Probability theory and related fields\/ 74\/ (2), 199--211

    1987

  36. [44]

    Glynn, A. N. and K. M. Quinn (2010). An introduction to the augmented inverse propensity weighted estimator. Political analysis\/ 18\/ (1), 36--56

    2010

  37. [45]

    Goodman, L. A. (1953). Ecological regressions and behavior of individuals. American sociological review\/ 18\/ (6), 663

    1953

  38. [46]

    Greiner, J. D. and K. M. Quinn (2009). R c ecological inference: bounds, correlations, flexibility and transparency of assumptions. Journal of the Royal Statistical Society Series A: Statistics in Society\/ 172\/ (1), 67--81

    2009

  39. [47]

    Hasminskii, R. and I. Ibragimov (2006). On asymptotic efficiency in the presence of an infinitedimensional nuisance parameter. In Probability Theory and Mathematical Statistics: Proceedings of the Fourth USSR-Japan Symposium, held at Tbilisi, USSR, August 23--29, 1982 , pp.\ 1...

    2006

  40. [48]

    Lu, and A

    Imai, K., Y. Lu, and A. Strauss (2008). Bayesian and likelihood inference for 2 2 ecological tables: an incomplete-data approach. Political Analysis\/ 16\/ (1), 41--69

    2008

  41. [49]

    Lei, and A

    Ji, W., L. Lei, and A. Spector (2023). Model-agnostic covariate-assisted inference on partially identified causal effects. arXiv preprint arXiv:2310.08115\/

    2023 arXiv

  42. [50]

    Kaji, T. and J. Cao (2023). Assessing heterogeneity of treatment effects. arXiv preprint arXiv:2306.15048\/

    2023 arXiv

  43. [51]

    Mao, and A

    Kallus, N., X. Mao, and A. Zhou (2022). Assessing algorithmic fairness with unobserved protected class using data combination. Management Science\/ 68\/ (3), 1959--1981

    2022

  44. [52]

    Kennedy, E. H. (2022). Semiparametric doubly robust targeted double machine learning: a review. arXiv preprint arXiv:2203.06469\/

    2022 arXiv

  45. [53]

    Kenny, C. T., S. Kuriwaki, C. McCartan, E. Rosenman, T. Simko, and K. Imai (2021). The impact of the us census disclosure avoidance system on redistricting and voting rights analysis. arXiv preprint arXiv:2105.14197\/

    2021

  46. [54]

    King, G. (2013). A solution to the ecological inference problem: Reconstructing individual behavior from aggregate data . Princeton University Press

    2013

  47. [55]

    King, G., M. A. Tanner, and O. Rosen (2004). Ecological inference: New methodological strategies . Cambridge University Press

    2004

  48. [56]

    Lee, D. S. (2005). Training, wages, and sample selection: Estimating sharp bounds on treatment effects

    2005

  49. [57]

    Li, S. and A. Luedtke (2023). Efficient estimation under data fusion. Biometrika\/ 110\/ (4), 1041--1054

    2023

  50. [58]

    Liu, S. and E. Dobriban (2019). Ridge regression: Structure, cross-validation, and sketching. arXiv preprint arXiv:1910.02373\/

    2019 arXiv

  51. [59]

    Makarov, G. (1982). Estimates for the distribution function of a sum of two random variables when the marginal distributions are fixed. Theory of Probability & its Applications\/ 26\/ (4), 803--806

    1982

  52. [60]

    Mankiw, N. G. (2013). Macroeconomics\/ (8th ed ed.). Worth Publishers

    2013

  53. [61]

    Manski, C. F. (2003). Partial identification of probability distributions , Volume 5. Springer

    2003

  54. [62]

    Manski, C. F. (2018a). Credible ecological inference for medical decisions with personalized risk assessment. Quantitative Economics\/ 9\/ (2), 541--569

    2018

  55. [63]

    Manski, C. F. (2018b). Credible ecological inference for medical decisions with personalized risk assessment. Quantitative Economics\/ 9\/ (2), 541--569

    2018

  56. [64]

    Marshall, R. J. (1990). Validation study methods for estimating exposure proportions and odds ratios with misclassified data. Journal of clinical epidemiology\/ 43\/ (9), 941--947

    1990

  57. [65]

    Molinari, F. (2008). Partial identification of probability distributions with misclassified data. Journal of Econometrics\/ 144\/ (1), 81--117

    2008

  58. [66]

    Muellera, J. T. and A. R. Santos-Lozadab. Proposed us census bureau differential privacy method is biased against rural and non-white. American Academy of Political and Social Science\/ 672\/ (1), 26--45

  59. [67]

    Pacini, D. (2019). Two-sample least squares projection. Econometric Reviews\/ 38\/ (1), 95--123

    2019

  60. [68]

    R \"a ssler, S. (2012). Statistical matching: A frequentist theory, practical applications, and alternative Bayesian approaches , Volume 168. Springer Science & Business Media

    2012

  61. [69]

    Robins, J. M., F. Hsieh, and W. Newey (1995). Semiparametric efficient estimation of a conditional density with missing or mismeasured covariates. Journal of the Royal Statistical Society: Series B (Methodological)\/ 57\/ (2), 409--424

    1995

  62. [70]

    Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of educational Psychology\/ 66\/ (5), 688

    1974

  63. [71]

    Fitch, D

    Ruggles, S., C. Fitch, D. Magnuson, and J. Schroeder (2019). Differential privacy and census data: Implications for social and economic research. In AEA papers and proceedings , Volume 109, pp.\ 403--408. American Economic Association 2014 Broadway, Suite 305, Nashville, TN 37203

    2019

  64. [72]

    Russell, T. M. (2021). Sharp bounds on functionals of the joint distribution in the analysis of treatment effects. Journal of Business & Economic Statistics\/ 39\/ (2), 532--546

    2021

  65. [73]

    Santos-Lozada, A. R., J. T. Howard, and A. M. Verdery (2020). How differential privacy will affect our understanding of health disparities in the united states. Proceedings of the National Academy of Sciences\/ 117\/ (24), 13405--13412

    2020

  66. [74]

    Stock, J. H., M. W. Watson, et al. (2003). Introduction to econometrics , Volume 104. Addison Wesley Boston

    2003

  67. [75]

    Armstrong, and P

    Wacholder, S., B. Armstrong, and P. Hartge (1993). Validation studies using an alloyed gold standard. American journal of epidemiology\/ 137\/ (11), 1251--1258

    1993

  68. [76]

    Wakefield, J. (2001). Ecological inference for 2 2 tables. Techincal Report, Department of Statistics and Biostatistics, University of Washington, USA\/

    2001

  69. [77]

    Berrendero, J. R. , Cuevas, A. & Torrecilla, J. L. (2016). On the use of reproducing kernel H ilbert spaces in functional classification. arXiv : 1507.04398v3

    2016 arXiv

  70. [78]

    Cox, D. R. (1972). Regression models and life tables (with Discussion) . J. R. Statist. Soc. B 34, 187--220

    1972

  71. [79]

    Heard, N. A. , Holmes, C. C. & Stephens, D. A. (2006). A quantitative study of gene regulation involved in the immune response of A nopheline mosquitoes: A n application of B ayesian hierarchical clustering of curves. J. Am. Statist. Assoc. 101, 18--29

    2006

  72. [80]

    R: A Language and Environment for Statistical Computing

    R Development Core Team (2024). R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0, http://www.R-project.org

    2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.