Testing conditional independence under isotonicity
Pith reviewed 2026-05-23 05:48 UTC · model grok-4.3
The pith
A swapping test for conditional independence of X and Y given Z controls Type I error exactly in finite samples when X is stochastically nondecreasing in Z.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The PairSwap-ICI procedure, which determines significance by randomly swapping the X values within ordered pairs of Z values where the matched pairs and test statistic may depend on both Y and Z, provides finite-sample Type I error control for testing conditional independence of X and Y given Z under the assumption that X is stochastically nondecreasing in Z, while provably achieving high power against a large class of alternatives.
What carries the argument
The PairSwap-ICI procedure that randomly swaps X values within ordered pairs of Z values, with pairs and statistic allowed to depend on Y and Z.
If this is right
- Analysts can test conditional independence in data sets where smoothness or parametric assumptions on the conditional distributions are implausible.
- The same swapping mechanism can be paired with any test statistic that incorporates information from Y and Z to target specific alternatives.
- Finite-sample validity holds without requiring large-sample approximations or knowledge of the marginal distribution of Z.
- Power is guaranteed against alternatives in which the dependence between X and Y given Z violates the monotonicity in a detectable way.
Where Pith is reading between the lines
- The method may be combined with existing rank-based or kernel statistics already used for unconditional independence testing.
- In observational studies where stochastic monotonicity of X in Z is plausible from domain knowledge, the test supplies a practical alternative to fully nonparametric procedures that lack finite-sample guarantees.
- Extensions could replace the pairwise swaps with larger blocks while preserving the exact control property under the same isotonicity condition.
Load-bearing premise
X is stochastically nondecreasing in Z.
What would settle it
A finite-sample simulation generated exactly under the null with the stochastic monotonicity holding, in which the empirical rejection rate of PairSwap-ICI exceeds the nominal alpha level.
read the original abstract
We propose a test of the conditional independence of random variables $X$ and~$Y$ given~$Z$ under the additional assumption that $X$ is stochastically nondecreasing in~$Z$. The well-documented hardness of testing conditional independence means that some further restriction on the null hypothesis parameter space is required. In contrast to existing approaches based on parametric models, smoothness assumptions, or approximations to the conditional distribution of $X$ given $Z$ and/or $Y$ given $Z$, our test requires only the stochastic monotonicity assumption. Our procedure, called \textnormal{\texttt{PairSwap-ICI}}, determines the significance of a statistic by randomly swapping the $X$ values within ordered pairs of~$Z$ values. The matched pairs and the test statistic may depend on both $Y$ and $Z$, providing the analyst with significant flexibility in constructing a powerful test. Our test offers finite-sample Type~I error control, and provably achieves high power against a large class of alternatives. We validate our theoretical findings through a series of simulations and real data experiments.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes the PairSwap-ICI procedure to test conditional independence of X and Y given Z under the additional restriction that X is stochastically nondecreasing in Z. The method constructs a null distribution by randomly swapping X values within ordered pairs of Z (with pairs and the test statistic permitted to depend on Y and Z), claiming finite-sample Type I error control together with provably high power against a broad class of alternatives; validity is illustrated via simulations and real-data examples.
Significance. If the finite-sample guarantee holds, the result supplies an exact nonparametric test for conditional independence that requires only stochastic monotonicity rather than parametric models or smoothness conditions. The explicit flexibility to let both the pairing rule and the test statistic depend on Y and Z is a practical strength that can be exploited to obtain power in applied settings where the isotonicity assumption is plausible.
minor comments (3)
- [Abstract] The abstract states that the matched pairs and test statistic 'may depend on both Y and Z,' but the precise conditions under which this dependence preserves the finite-sample Type I control should be stated explicitly in the main theorem (currently only alluded to).
- Notation for the ordered pairs and the swapping probability is introduced without a dedicated display equation; adding a short algorithmic box or numbered display would improve readability.
- The simulation section reports power curves but does not tabulate the empirical Type I error rates at the nominal level for the smallest sample sizes; a small table would directly confirm the finite-sample claim.
Simulated Author's Rebuttal
We thank the referee for their positive summary of our work on PairSwap-ICI and for recommending minor revision. The referee accurately captures the finite-sample validity under stochastic monotonicity and the flexibility in pairing and statistic construction. No specific major comments were listed in the report.
Circularity Check
No significant circularity identified
full rationale
The paper's central claim is finite-sample Type I error control for PairSwap-ICI under conditional independence plus the isotonicity assumption (X stochastically nondecreasing in Z). The swapping mechanism is constructed to generate an exact null distribution by exploiting exchangeability induced by those assumptions; the validity statement is therefore a direct consequence of the test design rather than a reduction to fitted parameters, self-definitions, or self-citations. No equations or steps in the abstract reduce a prediction to an input by construction, and the procedure is presented as a standard permutation-style test whose guarantees hold under the stated restrictions without circularity. The derivation chain is self-contained against external benchmarks.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption X is stochastically nondecreasing in Z
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.