Abstract

In many experiments the participant is presented with a stimulus and is required to make a binary judgement about it – for example whether they have encountered it before, whether it belongs to a category that they have previously learnt, or, more simply, whether or not it is present. These “yes-no” decisions are also required outside of the laboratory, for example when interpreting an X-ray trace for signs of disease or when sitting in the garden and deciding whether one really did hear the doorbell. In all of these situations one’s decision is influenced not only by the sensory information available but also by one’s criterion, or willingness to say “yes”. (Were you expecting a dearfriend, or your Ph.D. supervisor?). For the experimenter, this can complicate the interpretation of the results. In this lecture I will explain why commonly used measures of performance, such as percent correct, can be misleading, and explain how signal detection theory can help untangle the effects of criterion and of underlying sensitivity. I will also describe experimental methods that help circumvent this problem. Those attending should bring a calculator, although the maths involved will amount to no more than simple addition and subtraction.

For those keen students who would appreciate further discussion of experimental design and interpretation, I will make available a few examples of common logical and statistical errors that one encounters. Those interested will have the opportunity, at the end of the lecture, to discuss these examples and to propose ways of avoiding such pitfalls in one’s own research