Model: log k(year) = a + b·(year−1985)/10 · n=294 (220 complete, 74 censored) · effect measured against the null-distribution center (D-063), not against zero
The blue curve and band are the raw fit (b=0.092, 95% bootstrap interval, clustered by pair, B=2,000) — this is not the artifact-corrected effect from panel 2 (0.195); the two panels show different quantities. The orange dashed curve is what the artifact alone implies (b=−0.103, same anchor at 1985) — the "nothing is happening" baseline. Gray dashed: k=1, a memoryless process.
The 2020 interval runs from 0.801 to 1.013 and therefore includes k=1: by that year the data can no longer be distinguished from a memoryless process. It is not evidence that the parameter reached or passed 1 — the point estimate is 0.904. That boundary separates "conflicts stop clustering" from "a rhythm appears," and is the easiest thing to get wrong when reading this result.
| year | k (raw fit) | 95% CI |
|---|
Effect = b_obs − null-distribution center. Two null centers: the original one (§5, D-063) and the check of explanation B — year-dependent scale (D-068).
| quantity | value | 95% CI |
|---|
The effect against the alternative null (D-068, explanation-B check) has no formally computed combined interval — the interval shown here approximates it from the alternative null distribution's SD (±1.96·SD), the same way the main effect in protocol §4 is built.