A correction for independent Bernoulli trials reduced the estimated two-visit agreement gap between mean accuracy and the blink score in 40 ds001810 participants

We studied two attention scores across two stored visits in OpenNeuro ds001810 version 1.1.0. Forty people met the data rules. We estimated trial error from binary target responses. The model assumed independent trials with fixed response probabilities within each person, visit, and target delay. The correction reduced the estimated absolute agreement gap from 0.229 to 0.109. The decrease was 0.120, with a 95% bootstrap interval of 0.046 to 0.244. The corrected gap had an interval of -0.121 to 0.353. These intervals kept defined moment ratios. They do not show equal agreement. The corrected blink covariance matrix was invalid in 35.2% of 20,000 resamples. This limits the correction. We provide a proof for unequal trial counts, source events, file hashes, reusable code, and all results. A fresh public download reproduced all 39 checked result files.

attentional blink · test-retest agreement · intraclass correlation · trial error · Bernoulli model · bootstrap