Half of residual heat variation came from 22-39 monthly records in four Atlantic and Pacific coral samples in 2003-2019

We tested how much of a coral heat slope came from a small part of the survey record. Four Atlantic and Pacific samples from 2003-2019 had 5,336 monthly records at sites observed in two or more years. After control for site, year, month and depth, 22-39 records supplied half of squared residual heat variation in each sample. These records were 1.2-5.3 percent of each sample. The inverse sum of squared shares was 54-89 records. Removing the top five percent and fitting the model again changed slopes but left all four positive. This calculation measures design concentration. It does not show that high-share records are wrong or that they cause bias. Selected surveys, missing outcomes, few error clusters and one reporting system limit inference. The result does not measure species loss, pollution effects or a causal climate effect. Code, source hashes, survey evidence, checks and a figure are included.

coral bleaching · heat stress · observation bias · design concentration · fixed effects · regional replication