Effect of Response Bias on Threshold Estimates in Confounded N-Alternative Optional Choice Experiments

Three modifications of the familiar 2-AFC task can be useful when collecting data to estimate sensory thresholds from unpracticed participants: (1) Make responding optional, which relieves the participant from having to guess when they are uncertain; (2) Provide N>2 alternatives, which reduces the number of trials to estimate threshold; and (3) Confound the alternatives with an independent variable, for example by having participants choose visual field locations as alternatives when measuring threshold separately at each visual field location, which reduces the number of trials through multiplexing. The N-alternative optional-choice experiment (N-AOC) and its confounded variant have not been studied systematically. Confounded N-AOC has recently been adopted for use in clinical settings (e.g. Chia et al, Ophthalmology Glaucoma 2024) so it is important to understand its limitations. What happens when the participant prefers one response alternative to the others? How should this bias be modeled, what is its effect on threshold measurements, and can it be corrected for? Response biases occur for two reasons: (1) the criterion for responding is lower at one of the alternatives (N≥1), or (2) the lapse rate is high and the participant has a bias when guessing. Although catch (no-signal) trials are always included in yes-no experiments, they are not usually included in unconfounded N-AFC because the alternatives can be randomized. However, for confounded N-AFC and confounded N-AOC, catch trials must be included for the experimenter to detect or characterize response bias and to correct its pernicious effect on threshold estimates.