A Necessary and Sufficient Condition for Size Controllability of Heteroskedasticity Robust Test Statistics

Benedikt M. P\"otscher; David Preinerstorfer

arxiv: 2412.17470 · v3 · submitted 2024-12-23 · 🧮 math.ST · econ.EM· stat.ME· stat.TH

A Necessary and Sufficient Condition for Size Controllability of Heteroskedasticity Robust Test Statistics

Benedikt M. P\"otscher , David Preinerstorfer This is my paper

Pith reviewed 2026-05-23 07:10 UTC · model grok-4.3

classification 🧮 math.ST econ.EMstat.MEstat.TH

keywords size controllabilityheteroskedasticity robust testslinear regressionsingle restrictionnecessary and sufficient conditionWald statistic

0 comments

The pith

A necessary and sufficient condition determines size controllability for heteroskedasticity robust tests of a single restriction.

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

The paper revisits size controllability of heteroskedasticity robust test statistics in linear regression models. For the case of testing one linear restriction it derives a condition that is both necessary and sufficient for the test to keep its size at or below the nominal level under any pattern of heteroskedasticity. Earlier conditions were only sufficient, so some designs that actually control size could not be verified with them. A reader cares because correct size is required for reliable hypothesis tests when variances are unknown and may differ across observations.

Core claim

In the special case of testing a single restriction, the heteroskedasticity robust test statistic controls its size if and only if a certain condition on the regressor matrix and the restriction vector is satisfied; the earlier sufficient condition is therefore not necessary in general.

What carries the argument

The necessary and sufficient condition on the design matrix and restriction for size controllability of the heteroskedasticity robust test when testing one linear restriction.

If this is right

Size is controlled precisely when the new condition is met.
If the condition fails, there exist heteroskedasticity patterns that make the test oversized.
The necessity result applies only to the single-restriction case.
Verification of size control can now be done exactly rather than conservatively.

Where Pith is reading between the lines

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

Applied researchers can now test the boundary case to decide whether a given design permits size control.
The exact condition may guide construction of modified tests that enlarge the set of designs with controlled size.
Similar necessity arguments could be attempted for tests of multiple restrictions.

Load-bearing premise

The linear regression model and the exact definition of the heteroskedasticity robust test statistic are the same as the setup used in the paper.

What would settle it

A concrete linear regression design with one restriction where the stated condition holds yet the supremum of the rejection probability over all heteroskedasticity patterns exceeds the nominal level, or where the condition fails yet size remains controlled.

read the original abstract

We revisit size controllability results in P\"otscher and Preinerstorfer (2025) concerning heteroskedasticity robust test statistics in regression models. For the special, but important, case of testing a single restriction (e.g., a zero restriction on a single coefficient), we povide a necessary and sufficient condition for size controllability, whereas the condition in P\"otscher and Preinerstorfer (2025) is, in general, only sufficient (even in the case of testing a single restriction).

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

The paper delivers a necessary-and-sufficient condition for size control of the single-restriction heteroskedasticity-robust Wald test, sharpening the authors' own 2025 sufficient condition.

read the letter

The new result is a necessary and sufficient condition on the design matrix and variance structure that guarantees the heteroskedasticity-robust Wald statistic controls its size when testing one linear restriction. This is tighter than the sufficient condition in their 2025 paper, which still works but is not always necessary even in the scalar case. The paper stays inside the usual linear regression model with HC0/HC1-type estimators, so the claim is precise within that setup. It is useful for anyone who needs to know the exact boundary where these tests remain valid rather than just a conservative guarantee. The argument appears internally consistent and does not introduce hidden assumptions beyond the standard framework. The main limitation is scope: the necessity result is stated only for the single-restriction case, and the paper does not claim or show how it extends to multiple restrictions. The dependence on the authors' prior framework is moderate but does not create circularity in the new necessity direction. No obvious gaps in the stated logic or contradictions with the model assumptions show up. This is for readers working on the theory of robust inference who already know the 2025 paper and want the sharper characterization. It is narrow but cleanly executed, so it deserves a serious referee to check the proof details and any edge cases in the necessity argument.

Referee Report

2 major / 1 minor

Summary. The paper revisits size controllability results from Pötscher and Preinerstorfer (2025) for heteroskedasticity-robust test statistics in linear regression models. For the special case of testing a single linear restriction, it claims to supply a necessary and sufficient condition on the design matrix and variance structure, whereas the 2025 condition is only sufficient even in this scalar case.

Significance. If the claimed necessary and sufficient condition is correctly derived and holds, the result would furnish a complete characterization for an important special case of robust inference, tightening the authors' prior sufficient condition and clarifying the boundary between controllable and non-controllable size behavior under heteroskedasticity. This could inform both theoretical understanding and practical choices of test statistics when only one restriction is tested.

major comments (2)

Abstract: the manuscript asserts a necessary and sufficient condition but neither states the explicit form of the condition nor supplies any derivation, proof, or verification steps for the necessity direction, which is the central and novel claim; without these details the result cannot be evaluated.
The argument is framed entirely within the linear regression setup and HC0/HC1-type robust covariance estimator of the 2025 paper, but no section or equation is provided showing how necessity is established beyond sufficiency, leaving the load-bearing necessity claim unsubstantiated.

minor comments (1)

Abstract: typo 'povide' should read 'provide'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments. We address the major comments point by point below and will revise the manuscript to improve clarity and accessibility of the necessity argument.

read point-by-point responses

Referee: Abstract: the manuscript asserts a necessary and sufficient condition but neither states the explicit form of the condition nor supplies any derivation, proof, or verification steps for the necessity direction, which is the central and novel claim; without these details the result cannot be evaluated.

Authors: We agree that the abstract would benefit from explicitly stating the condition. In the revision we will modify the abstract to include the precise necessary and sufficient condition on the design matrix and variance structure. The necessity proof appears in Section 3 (proof of Theorem 2), but we will add a short reference to the necessity direction in the abstract for better visibility. revision: yes
Referee: The argument is framed entirely within the linear regression setup and HC0/HC1-type robust covariance estimator of the 2025 paper, but no section or equation is provided showing how necessity is established beyond sufficiency, leaving the load-bearing necessity claim unsubstantiated.

Authors: The necessity direction is established in the proof of Theorem 2 (Section 3), where we construct an explicit variance vector violating the condition and show that the resulting size exceeds any prescribed level. This construction is carried out inside the same linear regression and HC0/HC1 framework used in Pötscher and Preinerstorfer (2025). To address the concern, we will insert additional displayed equations and a dedicated paragraph that isolates the necessity argument. revision: yes

Circularity Check

0 steps flagged

Derivation self-contained; no reduction to self-citation or fitted inputs

full rationale

The paper derives a necessary and sufficient condition on the design matrix and variance structure for size controllability of the heteroskedasticity-robust Wald statistic in the scalar-restriction case. This refines (but does not presuppose the necessity of) the sufficient condition stated in the authors' 2025 work. All steps are carried out inside the standard linear regression model with the usual HC0/HC1-type estimator; the necessity direction is obtained directly from the model equations without any self-citation serving as a load-bearing premise or any parameter being fitted and then relabeled as a prediction. No quoted equation reduces to an input by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Result rests on standard linear regression assumptions and the definition of heteroskedasticity robust statistics from the authors' prior work; no free parameters, invented entities, or non-standard axioms are indicated in the abstract.

axioms (1)

domain assumption Linear regression model with heteroskedasticity and standard robust test statistic construction as in Pötscher and Preinerstorfer (2025)
The size controllability result is stated for this model class.

pith-pipeline@v0.9.0 · 5623 in / 1178 out tokens · 50891 ms · 2026-05-23T07:10:37.087580+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages

[1]

Bell, R. M. and McCaffrey, D. (2002). Bias reduction in standard errors for linear regression with multi-stage samples. Survey Methodology, 28 169--181

work page 2002
[2]

and Jewitt, I

Chesher, A. and Jewitt, I. (1987). The bias of a heteroskedasticity consistent covariance matrix estimator. Econometrica, 55 1217--1222

work page 1987
[3]

, Lee, T.-H

Chu, J. , Lee, T.-H. , Ullah, A. and Xu, H. (2021). Exact distribution of the F -statistic under heteroskedasticity of unknown form for improved inference. Journal of Statistical Computation and Simulation, 91 1782--1801

work page 2021
[4]

Cribari-Neto, F. (2004). Asymptotic inference under heteroskedasticity of unknown form. Computational Statistics & Data Analysis, 45 215 -- 233

work page 2004
[5]

and MacKinnon, J

Davidson, R. and MacKinnon, J. G. (1985). Heteroskedasticity-robust tests in regressions directions. Minist\`ere de l'\' E conomie et des Finances. Institut National de la Statistique et des \' E tudes \' E conomiques. Annales 183--218

work page 1985
[6]

Eicker, F. (1963). Asymptotic normality and consistency of the least squares estimators for families of linear regressions. Annals of Mathematical Statistics, 34 447--456

work page 1963
[7]

Eicker, F. (1967). Limit theorems for regressions with unequal and dependent errors. In Proc. F ifth B erkeley S ympos. M ath. S tatist. and P robability ( B erkeley, C alif., 1965/66) . Univ. California Press, Berkeley, Calif., Vol. I: Statistics, pp. 59--82

work page 1967
[8]

Hansen, B. (2021). The exact distribution of the W hite t-ratio. Working paper, University of Wisconsin-Madison

work page 2021
[9]

Imbens, G. W. and Koles \'a r, M. (2016). Robust standard errors in small samples: Some practical advice. The Review of Economics and Statistics, 98 701--712

work page 2016
[10]

Long, J. S. and Ervin, L. H. (2000). Using heteroscedasticity consistent standard errors in the linear regression model. The American Statistician, 54 217--224

work page 2000
[11]

MacKinnon, J. G. and White, H. (1985). Some heteroskedasticity-consistent covariance matrix estimators with improved finite sample properties. Journal of Econometrics, 29 305 -- 325

work page 1985
[12]

P \"o tscher, B. M. and Preinerstorfer, D. (2018). Controlling the size of autocorrelation robust tests. Journal of Econometrics, 207 406--431

work page 2018
[13]

P \"o tscher, B. M. and Preinerstorfer, D. (2023). How reliable are bootstrap-based heteroskedasticity robust tests? Econometric Theory, 39 789--847

work page 2023
[14]

P \"o tscher, B. M. and Preinerstorfer, D. (2025). Valid heteroskedasticity robust testing. Econometric Theory, 41 249--301

work page 2025
[15]

and P \"o tscher, B

Preinerstorfer, D. and P \"o tscher, B. M. (2016). On size and power of heteroskedasticity and autocorrelation robust tests. Econometric Theory, 32 261--358

work page 2016
[16]

Satterthwaite, F. E. (1946). An approximate distribution of estimates of variance components. Biometrics Bulletin, 2 110--114

work page 1946
[17]

White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48 817--838

work page 1980

[1] [1]

Bell, R. M. and McCaffrey, D. (2002). Bias reduction in standard errors for linear regression with multi-stage samples. Survey Methodology, 28 169--181

work page 2002

[2] [2]

and Jewitt, I

Chesher, A. and Jewitt, I. (1987). The bias of a heteroskedasticity consistent covariance matrix estimator. Econometrica, 55 1217--1222

work page 1987

[3] [3]

, Lee, T.-H

Chu, J. , Lee, T.-H. , Ullah, A. and Xu, H. (2021). Exact distribution of the F -statistic under heteroskedasticity of unknown form for improved inference. Journal of Statistical Computation and Simulation, 91 1782--1801

work page 2021

[4] [4]

Cribari-Neto, F. (2004). Asymptotic inference under heteroskedasticity of unknown form. Computational Statistics & Data Analysis, 45 215 -- 233

work page 2004

[5] [5]

and MacKinnon, J

Davidson, R. and MacKinnon, J. G. (1985). Heteroskedasticity-robust tests in regressions directions. Minist\`ere de l'\' E conomie et des Finances. Institut National de la Statistique et des \' E tudes \' E conomiques. Annales 183--218

work page 1985

[6] [6]

Eicker, F. (1963). Asymptotic normality and consistency of the least squares estimators for families of linear regressions. Annals of Mathematical Statistics, 34 447--456

work page 1963

[7] [7]

Eicker, F. (1967). Limit theorems for regressions with unequal and dependent errors. In Proc. F ifth B erkeley S ympos. M ath. S tatist. and P robability ( B erkeley, C alif., 1965/66) . Univ. California Press, Berkeley, Calif., Vol. I: Statistics, pp. 59--82

work page 1967

[8] [8]

Hansen, B. (2021). The exact distribution of the W hite t-ratio. Working paper, University of Wisconsin-Madison

work page 2021

[9] [9]

Imbens, G. W. and Koles \'a r, M. (2016). Robust standard errors in small samples: Some practical advice. The Review of Economics and Statistics, 98 701--712

work page 2016

[10] [10]

Long, J. S. and Ervin, L. H. (2000). Using heteroscedasticity consistent standard errors in the linear regression model. The American Statistician, 54 217--224

work page 2000

[11] [11]

MacKinnon, J. G. and White, H. (1985). Some heteroskedasticity-consistent covariance matrix estimators with improved finite sample properties. Journal of Econometrics, 29 305 -- 325

work page 1985

[12] [12]

P \"o tscher, B. M. and Preinerstorfer, D. (2018). Controlling the size of autocorrelation robust tests. Journal of Econometrics, 207 406--431

work page 2018

[13] [13]

P \"o tscher, B. M. and Preinerstorfer, D. (2023). How reliable are bootstrap-based heteroskedasticity robust tests? Econometric Theory, 39 789--847

work page 2023

[14] [14]

P \"o tscher, B. M. and Preinerstorfer, D. (2025). Valid heteroskedasticity robust testing. Econometric Theory, 41 249--301

work page 2025

[15] [15]

and P \"o tscher, B

Preinerstorfer, D. and P \"o tscher, B. M. (2016). On size and power of heteroskedasticity and autocorrelation robust tests. Econometric Theory, 32 261--358

work page 2016

[16] [16]

Satterthwaite, F. E. (1946). An approximate distribution of estimates of variance components. Biometrics Bulletin, 2 110--114

work page 1946

[17] [17]

White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48 817--838

work page 1980