Memory strength drives hybrid search efficiency

In hybrid search tasks, observers search for any of multiple targets held in memory (e.g., finding all of the ingredients for a recipe in the supermarket). Performance declines with the log of the number of memorized targets, a cost often attributed to a serial scan through memory. In standard models of recognition memory, however, holding more items in memory weakens the match strength between each memory representation and the corresponding visual item – independent of any serial search process. Here we propose an account of hybrid search based on memory strength, testing how memory strength affects hybrid search separately from memory set size. We found that strengthening memories through repetition and deeper encoding improved hybrid search at matched set sizes (Exp. 1). Taking advantage of NFL and NHL fans' pre-existing memory strength for sports logos (Exps. 2–3), we next found search was better for a larger set of familiar items (16 logos) than for a smaller set of unfamiliar items (12 logos), a reversal of the usual memory set-size effect. A standard global matching model (REM), with encoding parameters fit only to memory performance, not search data, reproduced these results and the canonical, approximately log-linear set-size cost in discriminability. Memory strength is thus a potentially strong determinant of hybrid search efficiency, with set-size effects not necessarily reflecting any serial memory search. Instead, models of recognition memory may provide a productive framework for the memory component of hybrid search.