For the WM paradigm, 9-s epochs (1.5 s pre-stimulus baseline of the first encoding item) were processed in an analogous fashion, although both correct and incorrect trials were included. Shorter 1.5-s subepochs (−102 to 1,398 ms around stimulus onset of each trial item) were extracted from these long, artifact-free ERPs using only the matching stimulus pairs (i.e., probe and corresponding encoding item), which were separately combined for each task. The means for the number of trials underlying these ERPs (M ± SD, HC vs. NH vs. AH) were 149 ±28 vs. 116 ±33 vs. 101 ±39 for faces, and 157 ±32 vs. 156 ±28 vs. 148 ±31 for words (range 62 to 202).
