Missing percentages permitted both heat-slope signs in three of four Atlantic and Pacific coral samples in 2003-2019

We tested whether missing coral bleaching percentages can change the sign of a heat slope. Four site, year, month, and depth models used 7,103 archived Reef Check records in the Atlantic and Pacific in 2003-2010 and 2011-2019. There were 623 missing percentages. If each missing percentage can take any value from 0 to 100, three exact coefficient ranges permit both signs. The later Atlantic range remains positive, from +1.453 to +1.812 percentage points per four TSA degree-heating weeks. Equal site-date weights and site-month controls give the same sign classification. Under equal cost per individual value, recovery of 8, 1, 48, and 36 values would remove at least half the respective bound widths. These are sharp ranges for fixed archive records, not confidence intervals or causal effects. Missing surveys, measurement error, uncertain dates, limited high heat, and unknown observer processes remain outside the bounds. Code, source hashes, an evidence frame, attainable endpoint assignments, and independent checks are included.

coral bleaching · heat stress · missing outcomes · partial identification · observation bias · Reef Check