In the previous version of this code there was a bug that affected the two-down one-up staircase, which when working properly should converge to approximately 71% correct (http://bdml.stanford.edu/twiki/pub/Haptics/DetectionThreshold/psychoacoustics.pdf). Briefly, a proper two-down one-up staircase should increase the signal strength one step when the subject makes an error ("one-up") and decrease it one step when there are two consecutive correct judgments ("two-down"). However the correct-trial counter should be reset to zero after two consecutive corrects, otherwise subsequent pairs are double-counted. E.g. if trial 1 is correct, and trial 2 is correct, then the task gets harder, but if trial 3 is correct then unless the counter is set to zero this will also be interpreted as a pair of correct trials (with trial 2). In simulation, this "modified" two-down one-up converges to a bit less than it should at ~65% correct (because it keeps making the task harder when it should keep it constant).
In previous versions of the code we also implemented an unequal stepsize which in combination with this modified rule and (again as verified by our simulations) does converge to ~70% correct. So this update should not change the convergence performance, but it now makes the staircase a true staircase, rather than a modified version with an unequal stepsize.
Summary
- updated bug in A/D key assignments
- fixed bug in staircase to implement true 2-down 1-up staircase with equal stepsize to converge to 71% correct