Balanced random halves gave a higher AX-CPT context score correlation than ordered file halves in 108 selected ds005754 v1.1.1 youths

We tested AX-CPT score consistency in public OpenNeuro ds005754 v1.1.1. We selected 108 participants with at least 70% correct BY trials and matching file identifiers. The corrected context score had a mean half correlation of 0.805 for 1,000 balanced random partitions. The correlation was 0.583 for the first and last 75 file positions. These file halves contained three and 12 BX trials. Random partitions with the same condition counts gave 0.771. A paired participant bootstrap gave a gap of 0.188, with a 95% interval of 0.092 to 0.286. Matching the condition counts did not remove the difference between random and ordered partitions. Two checks were selected after the first results. The score treats AX correct labels as hits and BX incorrect labels as false alarms. The source files do not show response types or trial times. The evidence is limited to one common task segment in this selected sample. It does not establish score stability across visits or a neural mechanism. The package preserves 117 source trial files, source hashes, code, scores, figures, bootstrap results, and reproduction instructions. A second complete analysis with newly retrieved source files gave the same results.

AX-CPT · split-half reliability · trial allocation · score consistency · individual differences · OpenNeuro