The Same‐Different Method: Positive Effects of Reduced Memory Load Versus Negative Effects of Uncontrolled τ‐Criterion Variation, Using Forced‐Choice Methods as a Comparison

The positive effects of reduced memory load for the same-different method were tested against the negative effects of judge τ-criterion variation, using the forced-choice triangle, duo-trio and ABX methods for comparison. The tests were carried out under nonlaboratory conditions in an everyday dining area, with mango juices as the stimuli. A control experiment involving sequential sensitivity analysis was performed to monitor carry-over effects. The results of the main experiment showed no significant differences in the low d′ values for the forced-choice methods, while a zero value was recorded for the same-different method. This was due to variability in the τ-criterion causing the overall value to be large enough to elicit, for sufficiently confusable stimuli, more incorrect than correct responses, when the two stimuli in the test were different. Practical Applications Sometimes it is required that difference tests be performed without the attribute change being described to the judges, leaving them to discover the change in attribute for themselves. In this case, the more statistically powerful methods, such as the 2-AFC and 3-AFC, are unsuitable. This leaves the much less powerful triangle, duo-trio and ABX methods as the most commonly used candidates, although their reduced power implies greater cost. Before the recent emergence of the statistically more powerful unspecified tetrad method as a potential candidate of choice, the long version of the same-different method was seen as the most powerful alternative. It has the advantage of a lesser memory load because only pairs of stimuli are assessed. However, the same-different method has the disadvantage of uncontrolled τ-criterion variation. There is a possibility that when data are pooled over a sample of consumers, the variance in τ-criteria among those judges might sometimes become large enough to cause problems. As long as the stimuli are sufficiently confusable, it might elicit a value large enough for incorrect responses to outnumber the correct responses, when the pairs in the test were different. The required Thurstonian analysis would then automatically record zero d′ values. Although this has not generally been regarded as a serious problem for the method, it would seem to be worthy of investigation. Such experiments are important for providing sensory professionals with information that enables them to make decisions regarding which methods are appropriate for their particular needs, when measuring discrimination.