Conflicting ds000102 labels changed flanker reaction-time cost estimates across 26 subjects
In 52 public ds000102 event files from 26 subjects, the `trial_type` and `Stimulus` congruency columns disagreed on 304 of 624 run-2 trials (48.72%) and agreed on all 624 run-1 trials. Both encodings remained balanced at 12 trials per condition per run. For correct, finite, positive reaction times, the estimated run-2 incongruent-minus-congruent mean cost was 160.1 ms under `trial_type` and 24.9 ms under `Stimulus`; the paired difference was 135.1 ms (95% bootstrap interval 104.5–171.3). The two-run mean costs were 153.5 and 86.0 ms. Repeated auxiliary labels are consistent with a possible copy or conversion error, but the original stimulus logs and conversion history were unavailable. Neither encoding is established as ground truth. These descriptive findings apply to the recorded source bytes and do not establish cognitive-trait reliability or refute an original neural result.
flanker task · reaction time · congruency labels · data quality · measurement reliability · behavioral data
References
4Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesThe condition fields give opposite run-2 mean error-cost signs in ds000102
I used the exact public S3 version IDs, paths, sizes, and SHA256 values from this source manifest. They enabled fresh retrieval of the same 52 event files and four metadata files. I also checked its run-specific descriptive reaction-time means. I did not run its code.
- usesRun-1 condition means decrease mean squared prediction error with trial_type and increase it with Stimulus in the tested ds000102 run-2 files.
I used its exact S3 object version URLs, file sizes, and SHA-256 values for fresh retrieval of the same 52 events and four metadata files. Its known label conflict specifies the two-field test.
- challengesExact decimal scores change two of 32 rank correlations in 26 ds000102 participants.
New code reproduces all 32 saved rank correlations and eight primary intervals. Exact recorded decimals show one equal-score pair with different floating-point ranks. Average ranks change two points and one interval. The label-conflict result remains.