Alternative designs

We have seen how signal detection theory offers a way in which to determine perceptual sensitivity from a yes/no task, separate from any influence of a bias towards responding "yes" or "no". Understanding the necessity of accounting for bias is fundamental to understanding psychophysical approaches to measuring perception.

However, the vast majority of vision research does not use the yes/no experimental design. Instead, they use alternative designs that are inherently less susceptible to observer bias.

The most common such experimental design is called a 'two-alternative forced-choice' (abbreviated as '2AFC'). The key difference is that, whereas a yes/no task presents only a single stimulus ('noise' or 'signal') on a given trial, the 2AFC method presents two stimuli on a given trial: one 'noise' and one 'signal'.

The two stimulus types in a 2AFC design can be presented simultaneously in different locations on the screen (on the left and the right, for example) and the observer is asked to judge the location of the 'signal' stimulus.

Left-click on the play icon () below to see an example:

Example trial using a spatial 2AFC design.

Alternatively, the two stimulus types can be presented at successive times (one second after each other, for example) and the observer is asked to judge whether the first or the second interval contained the 'signal' stimulus.

Left-click on the play icon () below to see an example:

Example trial using a temporal 2AFC design.

Why does this minimise the influence of bias?

To understand why the 2AFC approach is beneficial in terms of minimising the influence of bias, we can think through a 2AFC task in a signal detection framework. We will consider an example where the two stimuli in a trial are presented sequentially.

The presentation of the first stimulus will evoke some level of internal response in the observer. However, unlike the yes/no task, we do not require the observer to make a judgement based on this internal response. Instead, we present the second stimulus which will evoke another level of internal response in the observer. Now, the observer's task is to compare the magnitudes of internal response evoked by the two stimuli and respond with the interval that produced the larger response.

As you can see, there is no need to invoke a criterion in order to produce a judgement in a 2AFC task. It is this 'criterion-free' nature of 2AFC tasks that make such designs less susceptible to observer bias than yes/no tasks.

Estimating sensitivity using a 2AFC design is much simpler than with a yes/no design. Because we don't need to worry about disentangling sensitivity from bias, we can simply use observer accuracy as a measure of sensitivity. Accuracy is calculated by dividing the number of trials the observer responded correctly (e.g. chose the left side when the 'signal' was shown on the left and chose the right side when the 'signal' was shown on the right) by the total number of trials.

Summary

The susceptibility of yes/no tasks to observer bias has led researchers to prefer alternative experimental designs. A primary example is the two-alternative forced-choice design, in which both 'noise' and 'signal' stimuli are presented in each trial. This circumvents a yes/no bias, and allows sensitivity to be quantified via a straightforward accuracy measure (percentage correct).