Answers are grounded in this paper and cite their sources.
Background. Low-cost optical PM2.5 sensors such as PurpleAir overestimate concentrations and depend on relative humidity. Those distortions are largely monotone. Event triggers built on windowed Kendall τb have a proved, data-independent minimum inter-event time and are invariant under strictly increasing sensor maps, which makes them a natural candidate for episode detection on uncorrected low-cost series. A companion study in a Lorenz synchronisation loop found, however, that the same rules fire too late for tracking. Methods. A pre-registered confirmatory analysis asked whether the primary re-anchoring rule (R1), applied to uncorrected 24-hour PurpleAir PAcf1, recovers regulatory episodes at least as well as a magnitude detector on the same uncorrected series (H1) and at least as well as a magnitude detector on the published United States-wide EPA correction (H2). The data are the public collocated archive of Barkjohn, Gantt and Clements. The non-inferiority margin is Δ = 0.05 in event-level F1. Sites are the resampling units (M = 5000 bootstrap and permutation replicates; Holm adjustment at family-wise level 0.05). All design parameters were locked before episode-level scoring. Results. After completeness filtering, 37 sites and 9123 paired days were retained. The primary episode rule (PM25FM ≥ 35 µg m−3) produced only 8 episodes, so the pre-specified fallback—the site-level 90th percentile of the regulatory series—was invoked and became the sole definition used for scoring (519 episodes). Pooled F1 was 0.0285 for R1, 0.4109 for the raw magnitude rival and 0.7032 for the EPA-corrected rival. The differences were d1 = −0.3824 and d2 = −0.6747; both site-bootstrap 95% intervals lay entirely below −Δ, and both Holm-adjusted p-values were 1. Neither hypothesis was confirmed. Secondary ordinal rules did not improve on R1. A circular block-shift baseline was exceeded by both magnitude detectors and not by R1. Conclusion. The ordinal guarantees are not at issue. On these 24-hour series the re-anchoring trigger is a poor detector of short concentration spikes. It is better viewed as a structural alarm at the scale of its window, not as a substitute for a magnitude threshold or for the official correction.
No takes yet. Share an insight, caveat, or question.
Johel Padilla (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: