Item analysis in functional magnetic resonance imaging.

Behavioral and neuroimaging studies of cognition frequently test hypotheses regarding mental processing of different stimulus categories (e.g. verbs, faces, animals, scenes, etc.). The conclusions of such studies hinge upon the generalizability of their findings from the specific stimuli used in the experiment to the category as a whole. This type of generalizability is explicitly tested in behavioral studies, using "item analysis". However, generalizability to stimulus categories has up until now been assumed in neuroimaging studies, without employing item analysis for statistical validation. Here we apply item analysis to a functional magnetic resonance imaging study of nouns and verbs, demonstrating its theoretical importance and feasibility. In the subject-wise analysis, a left prefrontal and a left posterior-temporal region of interest showed putative grammatical class effects. An item-wise analysis revealed, however, that only the left posterior-temporal effect was generalizable to the stimulus categories of nouns and verbs. Taken together, the findings of the subject- and item-wise analyses suggest that grammatical-class effects in the left prefrontal cortex depend on the particular word stimuli used, rather than reflecting categorical differences between nouns and verbs. This empirical example illustrates that item analysis not only is sufficiently powered to detect task relevant changes in BOLD signal but also can make theoretically important distinctions between findings that generalize to the item populations, and those that do not.