Saberi K, Green D M
Department of Psychology, University of Florida, Gainesville 32611, USA.
J Acoust Soc Am. 1996 Jul;100(1):528-36. doi: 10.1121/1.415865.
One class of adaptive psychophysical procedures was studied, using simulated and human observers. These procedures are those which require an increase in stimulus intensity after an incorrect response, and a decrease after k successive correct responses. This paper analyzes how step size and the value of k affect the mean and standard deviation of threshold estimates based on a k-down 1-up adaptive procedure. Computer simulations are used to study the bias in threshold estimates, which are most evident when larger step size and small values of k are used. The adaptive procedure can be characterized by a function called the imbalance of the track, the relative probability of adjusting the stimulus either up or down at equal stimulus distances from the equilibrium point. These imbalance functions can be used to understand the threshold biases obtained in the computer simulations. The computer simulations also show that the average number of reversals obtained per trial is dependent on different values of k, but are largely independent of step size. The standard error of the threshold estimates, however, varies systematically with step size, but are nearly independent of k. Finally, we compare the stability of threshold estimates for human listeners using two very different sets of parameters: a very large step size (approximately half the range of the psychometric function) with k = 4, and the conventional k = 3 with an initial 4-dB and a final 2-dB step size.
使用模拟观察者和人类观察者，对一类自适应心理物理学程序进行了研究。这些程序是那些在错误反应后要求增加刺激强度，而在连续k次正确反应后要求降低刺激强度的程序。本文分析了步长和k值如何影响基于k降1升自适应程序的阈值估计的均值和标准差。使用计算机模拟来研究阈值估计中的偏差，当使用较大步长和较小k值时，这种偏差最为明显。自适应程序可以用一个称为轨迹不平衡的函数来表征，即在与平衡点相等刺激距离处向上或向下调整刺激的相对概率。这些不平衡函数可用于理解在计算机模拟中获得的阈值偏差。计算机模拟还表明，每次试验获得的平均反转次数取决于k的不同值，但在很大程度上与步长无关。然而，阈值估计的标准误差随步长系统地变化，但几乎与k无关。最后，我们使用两组非常不同的参数比较了人类听众阈值估计的稳定性：步长非常大（约为心理测量函数范围的一半）且k = 4，以及传统的k = 3，初始步长为4 dB，最终步长为2 dB。
