Power improvement from an educated guess of matching quality

Determine whether, in the absence of baseline data, using a pre-specified educated guess of the within-pair correlation parameter \(r\) in the optimally weighted estimator reliably improves power relative to the paired \(t\)-test.

Background

The optimal weight depends on the attrition rate and the within-pair correlation, which measures matching quality. When baseline outcomes are unavailable, the paper discusses estimating or prespecifying the correlation parameter instead of obtaining it directly from pretreatment data.

The authors conjecture that an educated guess based on pilot studies or similar experiments is likely to improve power over the paired tt-test, which implicitly treats matching as perfect by discarding singleton observations. The claim is presented as a conjecture rather than established theoretically or empirically in the paper.

References

We conjecture that even absent baseline data, a pre-specified educated guess for r (for instance, based on pilot studies or similar experiments) is likely to improve power over standard practice of using a paired t-test.

— Don't Drop the Singletons: Efficient Inference for Pairwise Experiments with Independent Attrition  (2608.18973 - Heß et al., 19 Aug 2026) in Section 2, paragraph “Feasible optimally weighted estimator”

We conjecture that a generalization of the dominance result (Proposition 7) holds under such heterogeneity, but leave this extension to future work.

— Don't Drop the Singletons: Efficient Inference for Pairwise Experiments with Independent Attrition  (2608.18973 - Heß et al., 19 Aug 2026) in Section 7, paragraph 2