Test order preserved odd and even study positions in 3,226 complete lists from OpenNeuro ds005059 version 1.0.6

All 3,226 complete lists in OpenNeuro ds005059 version 1.0.6 preserved odd and even study positions at test. The 36 observed test orders were exactly the permutations with this property. Study distance and test distance therefore always had the same parity. We also tested a planned model of 706 recall intrusions from 66 people. The base model included test order and word access. Adding linear study distance increased mean loss from 2.315014 to 2.321904 nats per error. Each person was predicted by a model fitted on other people. Gain was -0.006890, with a 2,000-sample person bootstrap interval of [-0.025093, 0.004793]. All fold models were refitted in each sample. All six planned checks had negative observed gains. Both primary models had greater observed loss than a uniform candidate distribution. The order constraint was an added descriptive check. The result concerns these records and fixed models. It does not establish a brain mechanism or the absence of temporal memory associations. The package preserves source bytes, hashes, code, candidate data, fits, bootstrap samples, results, and reproduction instructions.

paired-associate learning · cued recall · memory intrusions · study order · test order · cross-validation · person bootstrap · OpenNeuro