How do we evaluate evidence for the generality of phenomena in psycholinguistics and related fields? For decades, researchers have recognized the importance of treating not only participants, but also stimuli, as sampled rather than fixed (Clark, 1973). It is only by taking both subject and stimulus populations into account during analysis—and critically, doing so simultaneously—that claims about psycholinguistic phenomena are supported in their full generality. In this chapter, I argue that simultaneous by-subject and by-item analyses is a special (albeit common) case of a more general problem of generalizing over particular types of encounters, defined in terms of the sampled units involved in the situation and the connections between them. This approach is more general inasmuch as it encompasses the traditional two-party encounters between subjects and stimulus materials but also handles two-party dyadic communicative encounters, or even three-party encounters involving participants making judgments about stimuli that were produced by participants in response to other stimuli. I argue that the goal of analysis is to make claims that are maximally likely to generalize to new encounters of the same type, and propose that the relevant encounter types involved in an experiment can be identified by thinking about the boundaries of what would constitute a legitimate ‘replication’ of that experiment. These ideas have important implications far beyond psycholinguistics; indeed, nearly all areas of psychology and neuroscience make claims that are intended to generalize to new encounters, although it is only psycholinguistics that consistently uses the statistical apparatus requiredto support such claims.