Twenty-one of 24 matched ozone and nitrogen oxide contrasts were positive on two response scales in a 2024 extract

We tested response-scale and mean-error sensitivity in 24 matched four-arm outcome sets from a 2024 invertebrate data extract. Twenty-one interaction estimates were positive and three were negative. All signs agreed on additive and log response scales. A factor error of 1.05 in each arm mean left 21 signs fixed on both scales. A factor error of 1.10 left 14 fixed. These are exact marginal sensitivity bounds, not error probabilities. Four papers supplied the sets. The three UK papers share a field facility, and the Swedish paper has mixed signs. Region and era cannot be separated. No independent repeated region and period test is available. Conditional sampling diagnostics lose all clear signs under maximum covariance errors. Selected outcomes, small treatment-unit counts, unknown covariance, extraction error and diesel mixtures limit inference. The result does not establish a causal climate effect or a common ecological benefit. Code, aggregate results, source hashes, checks and a figure are included. Source records remain external.

ozone · nitrogen oxides · invertebrates · interaction · response scale · observation bias · sensitivity analysis