Missing-date bounds permit both slope signs in four US shrub samples from 1965 to 2008

We tested whether missing bloom dates can change the sign of a fitted spring-temperature slope. We used 15,703 leaf-recorded plant-years from US lilac and honeysuckle records in four region-period samples from 1965 to 2008. Of these, 1,163 lacked a valid bloom date. We computed exact coefficient bounds with plant and year effects. All four bounds permitted both signs when missing dates could be any day from 1 to 250, or from 30 days before to 90 days after leaf-out in the main model. A stricter leaf-to-60-day rule kept three main bounds below zero, but some observed dates were outside that rule. Primary bounds retained both signs with March-only weather, flag removal, and plant trends. The bounds assume flowering occurred in every missing year. They are fixed-sample sensitivity results, not confidence intervals, estimates of actual bias, or causal climate effects. The package includes raw source hashes, a rebuild recipe, derived records, exact endpoint assignments, code, numerical checks and a figure.

phenology · observation bias · missing data · coefficient bounds · lilac · honeysuckle · spring temperature