Pith. sign in

REVIEW 2 major objections 5 minor 133 references

Sign Hacking with Auxiliary Variable Exploration in the Age of Big Data

T0 review · 2 major / 5 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read In high-dimensional linear regression, a carefully chosen auxiliary variable can reverse targeted coefficient signs while making the results look statistically strong.

desk verdict Clean geometric existence result for selective sign reversal, plus a high-probability search theorem that formalizes a realistic high-dimensional p-hacking route; the dispersion assumption is the only real soft spot. read the letter →

arxiv 2607.04078 v1 pith:6BTEIZVT submitted 2026-07-05 stat.ME math.STstat.TH

Jie Hu , Yuhong Yang This is my paper
classification stat.MEmath.STstat.TH MSC 62J0562F0362P20
keywords signreversalp-hackingreproducibilitymodeluncertaintyresearchintegrityhigh-dimensionalregressionauxiliaryvariablesSHAVE
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

Coefficient signs tell researchers which way effects run, and they drive interpretation and policy. This paper shows that when many candidate covariates are available, a hacker can deliberately pick an auxiliary variable whose inclusion flips the signs of chosen coefficients while leaving others intact. The geometric set of such variables has positive volume in sample space, and under a dispersion condition on part of the auxiliary pool the search succeeds with probability approaching one as the pool grows. The same construction can also inflate t- and F-statistics, so the manipulated model looks significant and well-specified. Simulations and a gene-expression example confirm that the search is often easy in practice; the authors also sketch detection methods that use extra or independent data. The result matters because it supplies a concrete mechanism for p-hacking and for reproducibility failures that look statistically clean inside a single dataset.

What carries the argument

The positive-measure sign-change set V* constructed by V = XA + bY + εu with explicit bounds on A, b and ε (Theorem 2.1), which geometrically characterises all selective sign reversals and is then searched over under the conditional-independence Assumption 1.

What would settle it

Generate a large auxiliary pool that violates the dispersion condition (for example, all candidates highly collinear with the original X), then check whether the frequency of successful sign-reversing variables stays near zero even as pool size grows, contrary to Theorems 3.1–3.2.

Watch

Extended reading notes

Core claim

Conditional on the outcome and the regressors of interest, there is a measurable set of auxiliary-variable vectors of positive Lebesgue measure that reverse any prescribed subset of coefficient signs upon inclusion; under a mild conditional-independence dispersion condition on a subset of a large auxiliary pool, such a vector (and refinements that also produce large t- and F-statistics) is found with probability tending to one as the pool size grows.

Load-bearing premise

A subset of the auxiliary candidates must be conditionally independent given the outcome and main regressors and must spread out enough on the unit sphere; if the pool is tightly dependent or concentrated, the high-probability search claim fails.

Editorial extensions

If this is right

  • In high-dimensional empirical work, preferred coefficient signs can be manufactured by searching a rich enough auxiliary pool while still reporting significant t- and F-statistics.
  • Standard in-sample diagnostics (BIC, single-variable F-tests) can mechanically favour the hacked specification, so they offer little protection.
  • Lineup detection on augmented data and replication on independent samples become necessary credibility checks when model uncertainty is large.
  • Spurious sample correlations alone can reverse signs even when the auxiliary variable has no causal link to the outcome.

Reading between the lines

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

  • The same geometry implies that omitting a variable already inside V* can flip signs back, so both inclusion and exclusion searches are dual routes to the same manipulation.
  • Software that enumerates variable combinations or automates stepwise search lowers the practical cost of SHAVE to a few lines of code, making the theoretical possibility routinely attainable.
  • When baseline estimates already have the ‘wrong’ sign from sampling noise, a hacker can instead search for an auxiliary that preserves those signs while inflating significance, a quieter form of the same attack.
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 studies deliberate sign manipulation of OLS coefficients via inclusion of an auxiliary covariate drawn from a large pool, termed SHAVE. Conditional on (Y, X), Theorem 2.1 constructs an explicit positive-Lebesgue-measure set V of auxiliary realizations (of the form XA + bY + εu with sign restrictions on A, b and small ε) that reverse a prescribed subset of coefficients while preserving the rest. Under a conditional-independence-plus-uniform-density dispersion condition (Assumption 1) on a subset of size p1 of the auxiliary pool, Theorems 3.1–3.2 and Corollaries show that such a V (and refinements that also make the reversed coefficients and the auxiliary itself arbitrarily significant by t- and F-tests) is found with probability tending to one as p1 → ∞. Simulations with independent and constructed auxiliaries, plus a rat-eye gene-expression example, illustrate feasibility; lineup and replication detection strategies are proposed.

Significance. If the results hold, the paper supplies a clean geometric mechanism that converts high-dimensional model uncertainty into a concrete, searchable route for producing preferred sign patterns together with inflated significance. The constructive characterization of V (Eqs. 2.2–2.4) and the explicit reduction of classical sign-change conditions to vector angles are genuine strengths; the simulations and gene example make the finite-sample relevance tangible. The work usefully links sign instability, p-hacking, and research integrity, and the proposed detection ideas (lineup via selection frequency, replication with believability index τ) give empirical researchers something actionable. The contribution is therefore both theoretically clarifying and practically cautionary for high-dimensional applied work.

major comments (2)
  1. [§3, Assumption 1, Theorems 3.1–3.2] Assumption 1 and Theorems 3.1–3.2: the high-probability search-success claims rest entirely on the existence of a subset of p1 auxiliaries that are mutually independent given (Y, X) and whose normalized versions have densities uniformly bounded away from zero on the sphere. Real auxiliary pools (gene probes, macro indicators, text/image features) are typically highly dependent; while Remark 3.1 asserts that mild mixing can be accommodated, no rates, mixing coefficients, or modified covering arguments are supplied. A short extension or a clear quantitative caveat on when the probability may fail to approach one is needed for the abstract claim “with high probability … when many auxiliary candidates are available” to be operational.
  2. [§4.1, Table 2; Eq. (2.4)] Table 2 versus the construction of V (Eq. 2.4): under stronger correlation among the original regressors the empirical SHAVE frequency rises, yet the sufficient set V shrinks. The authors correctly note that V is only sufficient, but the discrepancy is left unresolved. Because the geometric insight of §2 is presented as the foundation for understanding when and why sign changes occur, a fuller (even partial) characterization of the necessary set, or at least a quantitative description of how dependence enlarges the true sign-change region, would make the theory more complete and would reconcile the simulation patterns with the constructive geometry.
minor comments (5)
  1. [Figure 1] Figure 1 caption and surrounding text: the figure is generated by repeated five-fold CV, yet the caption says “a single five-fold CV run” while the text refers to “repeated 10-fold CV”. Align the description with the actual experiment.
  2. [§2–3] Notation: the residual vectors are written both as êX and as ∥êX∥ > 0; later MX appears without re-definition in some places. A short notation paragraph or consistent use of the projection operators already introduced would help.
  3. [Appendix A] Appendix A, Lemma A.1 / Proposition A.1: the spherical-trigonometry identities are standard, but a one-line reference to the precise page of Palmer & Leigh (or a short derivation of the quadratic for sin η) would make the non-planar case self-contained.
  4. [§4.3, Table 4] Table 4 and the gene example: report the exact search procedure (exhaustive? random subsample? ordered by marginal correlation?) and the total number of candidates examined so that the “thousands of Vs” claim can be reproduced.
  5. [§5.2] Detection section: the believability index τ is introduced without guidance on how a practitioner should choose it; a short default recommendation or sensitivity plot would increase usability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: core theorems are constructive geometric/probabilistic existence results, not reductions of fitted inputs or self-referential definitions.

full rationale

The load-bearing claims (Theorem 2.1 existence of positive-Lebesgue-measure V* inducing selective sign reversals; Theorems 3.1–3.2 and Corollaries converting that set into high-probability search success under Assumption 1, plus t/F inflation) are derived from Euclidean geometry, orthogonal projections, and Frisch–Waugh–Lovell algebra. The set V is explicitly constructed as v = XA + b y + ε u with sign restrictions on A,b and small ε (Eqs. 2.2–2.4); membership forces the target sign pattern by linear-algebra identity, not by fitting a free parameter to the same quantity later called a prediction. Assumption 1 supplies dispersion so that independent normalized auxiliaries hit V with probability →1; this is an external regularity condition, not a definition of the target. Classical sign-change conditions are recovered as special cases of the same geometry (Appendix A), not smuggled uniqueness theorems. Self-citations (Nan–Yang 2014, Ye et al. 2018, Yang–Yang 2017/2025) appear only for motivation on selection instability and related p-hacking phenomena; none is invoked as a load-bearing uniqueness or existence premise for Theorems 2.1–3.2. Simulations and the gene-expression example are corroborative, not circular fits. The derivation chain is therefore self-contained against its own inputs.

Assumptions & free parameters 3 free parameters · 3 assumptions · 2 invented entities

The paper rests on standard Euclidean geometry and OLS algebra plus one non-trivial dispersion assumption that turns existence into high-probability search. Free parameters appear only in simulations and detection thresholds; the invented entity is the named practice SHAVE itself. No new physical or statistical objects beyond the geometric set V are postulated.

free parameters (3)
  • η (perturbation fraction) = 0.24
    Fixed at 0.24 in the sufficiency simulations (Section 4.2) to define the admissible ε-ball; chosen by hand inside (0,1/4).
  • c (t/F inflation multiple)
    Arbitrary positive constant used in Theorems 3.2 and corollaries to force |Tj| or FI above any multiple of the critical value; free design choice.
  • τ (believability index in replication detection)
    User-chosen threshold in (0,1) for the replication tests of Section 5.2; not estimated from data.
assumptions (3)
  • standard math Ordinary least-squares geometry: coefficient signs are determined by the orthogonal projection of Y onto the column space of the design; Frisch–Waugh–Lovell partialling applies.
    Used throughout Sections 2–3 and Appendix A to reduce sign questions to angles and parallelograms.
  • domain assumption Assumption 1 (Conditional Independence): a subset of p1 auxiliary candidates are mutually independent given (Y,X) and have densities bounded away from zero on the unit sphere.
    Load-bearing for the high-probability claims of Theorems 3.1–3.2; without it only the positive-measure existence result survives.
  • domain assumption Fixed finite q, full column rank of X, and non-zero residual norm (non-degenerate sample).
    Stated at the opening of Section 3; excludes singular or perfect-fit cases.
invented entities (2)
  • SHAVE (Sign Hacking with Auxiliary Variable Exploration)
    purpose: Names the deliberate (or exploratory) practice of searching a large auxiliary pool for a covariate that produces a desired sign pattern and inflated significance.
    Terminological framing; the underlying geometric set V is the mathematical object, not a new physical entity. No independent evidence required beyond the theorems themselves.
  • Constructed set V (and refinements Vδ,c, V1δ,c, Vδ,I,c) independent evidence
    purpose: Explicit positive-measure collection of auxiliary vectors that induce selective sign reversals (and optionally large t/F statistics).
    Defined constructively in Eqs. (2.2)–(2.4) and subsequent refinements; existence is proved, not postulated. Independent evidence is the geometric argument itself.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sign Hacking with Auxiliary Variable Exploration in the Age of Big Data." pith.science (2026). https://pith.science/paper/6BTEIZVT

@misc{pith2026260704078,
  author       = {Pith},
  title        = {Pith review of: Sign Hacking with Auxiliary Variable Exploration in the Age of Big Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6BTEIZVT}},
  note         = {Machine review of arXiv:2607.04078}
}
abstract

In linear regression, the signs of coefficients convey the direction of covariate effects and are central to empirical interpretation. In high-dimensional settings, however, the abundance of candidate covariates introduces substantial model selection uncertainty. We study the deliberate manipulation of coefficient signs through the inclusion of a carefully chosen auxiliary variable, a practice we term SHAVE (\textit{Sign Hacking with Auxiliary Variable Exploration}). We show that, conditional on the outcome and variables of interest, there exists a set of auxiliary-variable realizations with positive Lebesgue measure that lead to sign reversals upon inclusion. Moreover, with high probability, such variables can be found when many auxiliary candidates are available, leading simultaneously to reversed signs, inflated $t$- and $F$-statistics. Simulation studies and an empirical application corroborate these theoretical findings. We further propose detection strategies for SHAVE when augmented or independent datasets are available, as SHAVE has important implications for reproducibility, $p$-hacking, and research integrity.

Figures

Figures reproduced from arXiv: 2607.04078 by the authors.

Figure 1
Figure 1. Model selection uncertainty in the body fat dataset. Each row denotes a covariate, and each [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Geometric configurations for sign changes of [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. reports, for each control set size, both the number of candidate variables V and the corresponding number of control variable combinations. Across all configurations, multiple variables V consistently induce sign reversals, with a mild decline as the control set size increases. Moreover, each such variable typically operates across multiple specifications with the same control size. (a) Single XI1 case (b) Two-varia… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

133 extracted references

  1. [1]

    John Aldrich , year = 1995, journal =

    1995

  2. [2]

    Amini, Shahram M and Parmeter, Christopher F , year = 2012, journal =

    2012

  3. [3]

    Arcidiacono , year = 2018, howpublished =

    Peter S. Arcidiacono , year = 2018, howpublished =

    2018

  4. [4]

    David Card , year = 2018, howpublished =

    2018

  5. [5]

    Z. D. Bai and Y. Q. Yin , title =. The Annals of Probability , number =

  6. [6]

    JAMA , publisher =

    Why review articles on the health effects of passive smoking reach different conclusions , author =. JAMA , publisher =

  7. [7]

    Beasley, Malcolm R , year = 2002, howpublished =

    2002

  8. [8]

    Bernoulli , publisher =

    Least squares after model selection in high-dimensional sparse models , author =. Bernoulli , publisher =

Show all 133 references
  1. [9]

    PLoS Medicine , publisher =

    Financial conflicts of interest and reporting bias regarding the association between sugar-sweetened beverages and weight gain: a systematic review of systematic reviews , author =. PLoS Medicine , publisher =

  2. [10]

    JAMA , publisher =

    Scope and impact of financial conflicts of interest in biomedical research: a systematic review , author =. JAMA , publisher =

  3. [11]

    Lecture Notes-Monograph Series , publisher =

    An alternative point of view on Lepski's method , author =. Lecture Notes-Monograph Series , publisher =

  4. [12]

    American Economic Review , volume = 110, number = 11, pages =

    Methods matter: p-hacking and publication bias in causal analysis in economics , author =. American Economic Review , volume = 110, number = 11, pages =

  5. [13]

    Biometrics , publisher =

    Model selection: an integral part of inference , author =. Biometrics , publisher =

  6. [14]

    Journal of Econometrics , publisher =

    Toward optimal model averaging in regression models with time series errors , author =. Journal of Econometrics , publisher =

  7. [15]

    Chiang, Annie P and Beck, John S and Yen, Hsan-Jan and Tayeh, Marwan K and Scheetz, Todd E and Swiderski, Ruth E and Nishimura, Darryl Y and Braun, Terry A and Kim, Kwang-Youn A and Huang, Jian and others , year = 2006, journal =

    2006

  8. [16]

    Journal of Economic Literature , volume = 56, number = 3, pages =

    Transparency, reproducibility, and the credibility of economics research , author =. Journal of Economic Literature , volume = 56, number = 3, pages =

  9. [17]

    American Economic Journal: Macroeconomics , volume = 2, number = 4, pages =

    Determinants of economic growth: will data tell? , author =. American Economic Journal: Macroeconomics , volume = 2, number = 4, pages =

  10. [18]

    Journal of Economic Surveys , publisher =

    The meaning of failed replications: A review and proposal , author =. Journal of Economic Surveys , publisher =

  11. [19]

    Nature News , volume = 2, pages =

    Papers on ‘stress-induced’stem cells are retracted , author =. Nature News , volume = 2, pages =

  12. [20]

    AMS Math Challenges Lecture , publisher =

    High-dimensional data analysis: The curses and blessings of dimensionality , author =. AMS Math Challenges Lecture , publisher =

  13. [21]

    , year = 2004, journal =

    Sala-I-Martin, Xavier and Doppelhofer, Gernot and Miller, Ronald I. , year = 2004, journal =

    2004

  14. [22]

    Biomolecular Engineering , publisher =

    Fabrication of high quality microarrays , author =. Biomolecular Engineering , publisher =

  15. [23]

    Science , publisher =

    Economics in the age of big data , author =. Science , publisher =

  16. [24]

    Econometrica , volume=

    The model confidence set , author=. Econometrica , volume=. 2011 , publisher=

    2011

  17. [25]

    Journal of the American Statistical Association , volume=

    Cross-validation with confidence , author=. Journal of the American Statistical Association , volume=. 2020 , publisher=

    2020

  18. [26]

    Econometrica , volume = 90, number = 2, pages =

    Detecting p-Hacking , author =. Econometrica , volume = 90, number = 2, pages =

  19. [27]

    Journal of Business Research , publisher =

    Big Data consumer analytics and the transformation of marketing , author =. Journal of Business Research , publisher =

  20. [28]

    Journal of the American Statistical Association , publisher =

    Variable selection via nonconcave penalized likelihood and its oracle properties , author =. Journal of the American Statistical Association , publisher =

  21. [29]

    Journal of the American Statistical Association , publisher =

    Goodness-of-fit tests for parametric regression models , author =. Journal of the American Statistical Association , publisher =

  22. [30]

    Journal of the Royal Statistical Society: Series B (Statistical Methodology) , publisher =

    Sure independence screening for ultrahigh dimensional feature space , author =. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , publisher =

  23. [31]

    Statistica Sinica , publisher =

    A selective overview of variable selection in high dimensional feature space , author =. Statistica Sinica , publisher =

  24. [32]

    Journal of the Royal Statistical Society: Series B (Statistical Methodology) , publisher =

    Variance estimation using refitted cross-validation in ultrahigh dimensional regression , author =. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , publisher =

  25. [33]

    National Science Review , publisher =

    Challenges of big data analysis , author =. National Science Review , publisher =

  26. [34]

    Benchmark priors revisited: on adaptive shrinkage and the supermodel effect in Bayesian model averaging , author =

  27. [35]

    The impact of data revisions on the robustness of growth determinants—

    Feldkircher, Martin and Zeugner, Stefan , year = 2012, journal =. The impact of data revisions on the robustness of growth determinants—

    2012

  28. [36]

    Confidence sets for model selection by

    Ferrari, Davide and Yang, Yuhong , year = 2015, journal =. Confidence sets for model selection by

    2015

  29. [37]

    The Journal of Finance , publisher =

    Spurious regressions in financial economics? , author =. The Journal of Finance , publisher =

  30. [38]

    Journal of Investment Management , volume = 1, number = 3, pages =

    Is stock return predictability spurious , author =. Journal of Investment Management , volume = 1, number = 3, pages =

  31. [39]

    , author =

    Statistical methods for research workers. , author =

  32. [40]

    the American Statistician , publisher =

    A note on screening regression equations , author =. the American Statistician , publisher =

  33. [41]

    Gelman, Andrew and Loken, Eric , year = 2013, journal =

    2013

  34. [42]

    Econometrica , publisher =

    Measuring group differences in high-dimensional choices: method and application to congressional speech , author =. Econometrica , publisher =

  35. [43]

    Prevalence of questionable research practices, research misconduct and their potential explanatory factors: a survey among academic researchers in

    Gopalakrishna, Gowri and ter Riet, Gerben and Cruyff, Maarten JLF and Vink, Gerko and Stoop, Ineke and Wicherts, Jelte and Bouter, Lex , year = 2022, journal =. Prevalence of questionable research practices, research misconduct and their potential explanatory factors: a survey...

    2022

  36. [44]

    Journal of Econometrics , publisher =

    Spurious regressions in econometrics , author =. Journal of Econometrics , publisher =

  37. [45]

    Econometrica , publisher =

    Least squares model averaging , author =. Econometrica , publisher =

  38. [46]

    Journal of Econometrics , publisher =

    Jackknife model averaging , author =. Journal of Econometrics , publisher =

  39. [47]

    IEEE Transactions on Automation Science and Engineering , volume = 12, number = 1, pages =

    Image-Based Process Monitoring Using Low-Rank Tensor Decomposition , author =. IEEE Transactions on Automation Science and Engineering , volume = 12, number = 1, pages =

  40. [48]

    PLoS Biology , publisher =

    The extent and consequences of p-hacking in science , author =. PLoS Biology , publisher =

  41. [49]

    Journal of the American Statistical Association , publisher =

    Frequentist model average estimators , author =. Journal of the American Statistical Association , publisher =

  42. [50]

    Statistical Science , publisher =

    Bayesian model averaging: a tutorial , author =. Statistical Science , publisher =

  43. [51]

    Matrix analysis , author =

  44. [52]

    Journal of Statistics Education , publisher =

    Fitting percentage of body fat to simple body measurements , author =. Journal of Statistics Education , publisher =

  45. [53]

    Statistica Sinica , publisher =

    Adaptive Lasso for sparse high-dimensional regression models , author =. Statistica Sinica , publisher =

  46. [54]

    On the Smallest Eigenvalue of General correlated Gaussian Matrices , author =

  47. [55]

    JAMA Internal Medicine , publisher =

    Sugar industry and coronary heart disease research: a historical analysis of internal industry documents , author =. JAMA Internal Medicine , publisher =

  48. [56]

    The Advanced Theory of Statistics , author =

  49. [57]

    Kennedy, Peter E , year = 2005, journal =. Oh no!

    2005

  50. [58]

    The American Statistician , publisher =

    Reversals of Least-Square Estimates and Model-Invariant Estimation for Directions of Unique Effects , author =. The American Statistician , publisher =

  51. [59]

    Harvard calls for retraction of dozens of studies by noted cardiac researcher , author =

  52. [60]

    Political Analysis , publisher =

    Achieving statistical significance with control variables and without transparency , author =. Political Analysis , publisher =

  53. [61]

    Procedia Computer Science , publisher =

    L. Procedia Computer Science , publisher =

  54. [62]

    Duke University pays \ 112.5 million in fake research case sparked by whistleblower , author =

  55. [63]

    Journal of Econometrics , publisher =

    A result on the sign of restricted least-squares estimates , author =. Journal of Econometrics , publisher =

  56. [64]

    The American Economic Review , publisher =

    Let's take the con out of econometrics , author =. The American Economic Review , publisher =

  57. [65]

    PLoS Medicine , publisher =

    Relationship between funding source and conclusion among nutrition-related scientific articles , author =. PLoS Medicine , publisher =

  58. [66]

    IEEE Transactions on Information Theory , publisher =

    Information theory and mixing least-squares regressions , author =. IEEE Transactions on Information Theory , publisher =

  59. [67]

    The American Economic Review , publisher =

    A sensitivity analysis of cross-country growth regressions , author =. The American Economic Review , publisher =

  60. [68]

    Statistics in Medicine , publisher =

    Robust group variable screening based on maximum Lq-likelihood estimation , author =. Statistics in Medicine , publisher =

  61. [69]

    Journal of the American Statistical Association , publisher =

    Mixtures of g priors for Bayesian variable selection , author =. Journal of the American Statistical Association , publisher =

  62. [70]

    Journal of the American Statistical Association , publisher =

    Optimal weight choice for frequentist model average estimators , author =. Journal of the American Statistical Association , publisher =

  63. [71]

    Journal of Computational and Graphical Statistics , publisher =

    Estimation stability with cross-validation (ESCV) , author =. Journal of Computational and Graphical Statistics , publisher =

  64. [72]

    Stephen Lindsay , year = 2015, journal =

    D. Stephen Lindsay , year = 2015, journal =

    2015

  65. [73]

    Lovell , year = 1983, journal =

    Michael C. Lovell , year = 1983, journal =

    1983

  66. [74]

    Journal of Econometrics , publisher =

    Robustness checks and robustness tests in applied economics , author =. Journal of Econometrics , publisher =

  67. [75]

    A Hierarchy of Limitations in Machine Learning , author =

  68. [76]

    stargazer: Well-Formatted Regression and Summary Statistics Tables , author =

  69. [77]

    American Economic Review , year =

    The Effect of Omitted Variables on the Sign of Regression Coefficients , author =. American Economic Review , year =

  70. [78]

    The American Economic Review , publisher =

    What will take the con out of econometrics? , author =. The American Economic Review , publisher =

  71. [79]

    Journal of Econometrics , publisher =

    A further result on the sign of restricted least-squares estimates , author =. Journal of Econometrics , publisher =

  72. [80]

    Journal of the Royal Statistical Society: Series B (Statistical Methodology) , publisher =

    Stability selection , author =. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , publisher =

  73. [81]

    Journal of Economic Perspectives , volume = 31, number = 2, pages =

    Machine learning: an applied econometric approach , author =. Journal of Economic Perspectives , volume = 31, number = 2, pages =

  74. [82]

    Journal of Computational and Graphical Statistics , publisher =

    Variable selection diagnostics measures for high-dimensional regression , author =. Journal of Computational and Graphical Statistics , publisher =

  75. [83]

    Oxford Bulletin of Economics and Statistics , publisher =

    PRACTITIONERS’CORNER†: On Sign Changes Upon Deletion of a Variable in Linear Regression Analysis , author =. Oxford Bulletin of Economics and Statistics , publisher =

  76. [84]

    Journal of Business & Economic Statistics , volume=

    Unobservable selection and coefficient stability: Theory and evidence , author=. Journal of Business & Economic Statistics , volume=. 2019 , publisher=

    2019

  77. [85]

    Plane and spherical trigonometry , author =

  78. [86]

    The American Statistician , publisher =

    Understanding Simpson’s paradox , author =. The American Statistician , publisher =

  79. [87]

    Journal of the American Statistical Association , publisher =

    Bayesian model averaging for linear regression models , author =. Journal of the American Statistical Association , publisher =

  80. [88]

    R: A Language and Environment for Statistical Computing , author =

  81. [89]

    Computational Statistics & Data Analysis , publisher =

    Improving the robustness of fractional polynomial models by preliminary covariate transformation: A pragmatic approach , author =. Computational Statistics & Data Analysis , publisher =

  82. [90]

    American Economic Review , publisher =

    Determinants of long-term growth: A Bayesian averaging of classical estimates (BACE) approach , author =. American Economic Review , publisher =

  83. [91]

    Proceedings of the National Academy of Sciences , publisher =

    Regulation of gene expression in the mammalian eye and its relevance to eye disease , author =. Proceedings of the National Academy of Sciences , publisher =

  84. [92]

    Nature , volume = 563, number = 7733, pages =

    Statistical pitfalls of personalized medicine , author =. Nature , volume = 563, number = 7733, pages =

  85. [93]

    Shahbaz, Muhammad and Sinha, Avik , year = 2019, journal =

    2019

  86. [94]

    IEEE Signal Processing Magazine , publisher =

    The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains , author =. IEEE Signal Processing Magazine , publisher =

  87. [95]

    Psychological Science , publisher =

    False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant , author =. Psychological Science , publisher =

  88. [96]

    Journal of Experimental Psychology: General , publisher =

    P-curve: a key to the file-drawer , author =. Journal of Experimental Psychology: General , publisher =

  89. [97]

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

    The interpretation of interaction in contingency tables , author =. Journal of the Royal Statistical Society: Series B (Methodological) , publisher =

  90. [98]

    On the effect of prior assumptions in Bayesian model averaging with applications to growth regression , author =

  91. [99]

    Journal of the American statistical association , volume=

    Publication decisions and their possible effects on inferences drawn from tests of significance—or vice versa , author=. Journal of the American statistical association , volume=. 1959 , publisher=

    1959

  92. [100]

    Annual Review of Statistics and Its Application , publisher =

    Reproducing statistical results , author =. Annual Review of Statistics and Its Application , publisher =

  93. [101]

    Journal of Economic Literature , volume = 49, number = 2, pages =

    What determines productivity? , author =. Journal of Economic Literature , volume = 49, number = 2, pages =

  94. [102]

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

    Regression shrinkage and selection via the lasso , author =. Journal of the Royal Statistical Society: Series B (Methodological) , publisher =

  95. [103]

    Journal of Econometrics , publisher =

    On obtaining the right sign of a coefficient estimate by omitting a variable from the regression , author =. Journal of Econometrics , publisher =

  96. [104]

    Oxford Bulletin of Economics and Statistics , publisher =

    Again on sign changes upon deletion of a variable from a linear regression , author =. Oxford Bulletin of Economics and Statistics , publisher =

  97. [105]

    Journal of Econometrics , volume=

    Least squares model averaging by Mallows criterion , author=. Journal of Econometrics , volume=. 2010 , publisher=

    2010

  98. [106]

    Sharp thresholds for High-Dimensional and noisy sparsity recovery using

    Wainwright, Martin J , year = 2009, journal =. Sharp thresholds for High-Dimensional and noisy sparsity recovery using

    2009

  99. [107]

    Statistica Sinica , publisher =

    Robust combination of model selection methods for prediction , author =. Statistica Sinica , publisher =

  100. [108]

    Science and Engineering Ethics , publisher =

    Prevalence of research misconduct and questionable research practices: a systematic review and meta-analysis , author =. Science and Engineering Ethics , publisher =

  101. [109]

    IEEE Transactions on Information Theory , publisher =

    An asymptotic property of model selection criteria , author =. IEEE Transactions on Information Theory , publisher =

  102. [110]

    Journal of the American Statistical Association , publisher =

    Adaptive regression by mixing , author =. Journal of the American Statistical Association , publisher =

  103. [111]

    The Annals of Statistics , publisher =

    Consistency of cross validation for comparing regression procedures , author =. The Annals of Statistics , publisher =

  104. [112]

    Biometrics , publisher =

    Toward an objective and reproducible model choice via variable selection deviation , author =. Biometrics , publisher =

  105. [113]

    Journal of the American Statistical Association , publisher =

    Sparsity oriented importance learning for high-dimensional linear regression , author =. Journal of the American Statistical Association , publisher =

  106. [114]

    Journal of the Royal Statistical Society , publisher =

    An investigation into the causes of changes in pauperism in England, chiefly during the last two intercensal decades (Part I.) , author =. Journal of the Royal Statistical Society , publisher =

  107. [115]

    Why do we sometimes get nonsense-correlations between

    Yule, G Udny , year = 1926, journal =. Why do we sometimes get nonsense-correlations between

    1926

  108. [116]

    Zeugner, Stefan and Feldkircher, Martin and others , year = 2015, journal =

    2015

  109. [117]

    The Annals of Statistics , publisher =

    Nearly unbiased variable selection under minimax concave penalty , author =. The Annals of Statistics , publisher =

  110. [118]

    Journal of Machine Learning Research , volume = 7, pages =

    On model selection consistency of Lasso , author =. Journal of Machine Learning Research , volume = 7, pages =

  111. [119]

    Statistica Sinica , publisher =

    Model selection confidence sets by likelihood ratio testing , author =. Statistica Sinica , publisher =

  112. [120]

    Journal of the American statistical association , publisher =

    The adaptive lasso and its oracle properties , author =. Journal of the American statistical association , publisher =

  113. [121]

    Journal of the American Statistical Association , volume=

    Testing for polynomial regression using nonparametric regression techniques , author=. Journal of the American Statistical Association , volume=. 1996 , publisher=

    1996

  114. [122]

    Econometrica: Journal of the Econometric Society , pages=

    Consistent specification testing via nonparametric series regression , author=. Econometrica: Journal of the Econometric Society , pages=. 1995 , publisher=

    1995

  115. [123]

    Test , volume=

    An updated review of goodness-of-fit tests for regression models , author=. Test , volume=. 2013 , publisher=

    2013

  116. [124]

    Emerging themes in epidemiology , volume=

    Simpson’s Paradox is suppression, but Lord’s Paradox is neither: clarification of and correction to Tu, Gunnell, and Gilthorpe (2008) , author=. Emerging themes in epidemiology , volume=. 2019 , publisher=

    2008

  117. [125]

    Journal of the American Statistical Association , volume=

    Interaction screening for ultrahigh-dimensional data , author=. Journal of the American Statistical Association , volume=. 2014 , publisher=

    2014

  118. [126]

    Nature Reviews Genetics , volume=

    Detecting gene--gene interactions that underlie human diseases , author=. Nature Reviews Genetics , volume=. 2009 , publisher=

    2009

  119. [127]

    The Journal of Machine Learning Research , volume=

    On the stability of feature selection algorithms , author=. The Journal of Machine Learning Research , volume=. 2017 , publisher=

    2017

  120. [128]

    Journal of Marketing , volume=

    Attribute embedding: Learning hierarchical representations of product attributes from consumer reviews , author=. Journal of Marketing , volume=. 2022 , publisher=

    2022

  121. [129]

    Machine learning , volume=

    Bagging predictors , author=. Machine learning , volume=. 1996 , publisher=

    1996

  122. [130]

    The annals of statistics , volume=

    Heuristics of instability and stabilization in model selection , author=. The annals of statistics , volume=. 1996 , publisher=

    1996

  123. [131]

    Journal of Statistical Planning and Inference , volume=

    Model combining in factorial data analysis , author=. Journal of Statistical Planning and Inference , volume=. 2007 , publisher=

    2007

  124. [132]

    The New England Journal of Statistics in Data Science , volume=

    A Study on Reproducibility and the Reliability of the Hosmer-Lemeshow Test in Published Research , author=. The New England Journal of Statistics in Data Science , volume=. 2025 , publisher=

    2025

  125. [133]

    The annals of statistics , pages=

    Estimating the dimension of a model , author=. The annals of statistics , pages=. 1978 , publisher=

    1978

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.