Intervals with no common date had negative covariance estimates in four western US lilac samples in 1967-2003

We tested whether a negative link between spring development intervals persists when they share no date. We used 3,872 common lilac unit-years in two western US regions and two periods from 1967 to 2003. Models removed unit effects, year effects and spring temperature. Covariance between leaf-to-first-bloom and full-to-end-bloom intervals ranged from -9.68 to -5.38 days squared. Early-period marginal 95% bootstrap intervals were below zero. Both later-period intervals included zero. All estimates remained negative without weather, with separate monthly weather, after flag removal, after zero-interval removal and after each state omission. A direct shared date cannot alone explain the disjoint-interval calculation. Correlated reporting error, selection and unmeasured conditions remain possible causes. Few state clusters and missing visit records limit inference. This result does not establish biological compensation or a causal climate effect. The package gives four-stage evidence, code, source hashes, uncertainty and numerical checks.

phenology · lilac · observation bias · shared date error · development intervals · regional replication · uncertainty