Geoff Loftus: On the benefits of bogus data

Abstract

The responses in a psychological experiment traditionally come from living subjects drawn from whatever population is under investigation. Another potential source of responses is a computer program designed to mimic the subjects' responses. While one would not generally want to use these computer-generated or "bogus" responses as the ultimate basis for making conclusions about psychological processes, they are nevertheless useful in three respects. The first is methodological: One can test out almost all aspects of the planned experiment (e.g., computer programs that run the experiment and analyze the data from it) without the expense, time, and hassle of using actual subjects. The second is statistical: One can simulate various kinds of randomization and counterbalancing and observe the consequences for the orderliness of the data. The third is theoretical: To generate bogus data, one is forced to generate a theory of how the responses are produced and how they are affected by the various experimental manipulations. The process of so doing is often accompanied by remarkable insights into the nature of possible theories of the processes under investigation, along with immediate feedback about detailed predictions of such theories.