Percent Correct Measure Research Articles

Perception of ‘‘asymmetrical’’ German vowels by humans, monkeys and gerbils

Polka and Bohn [J. Acoust. Soc. Am. 100, 557–592 (1996)] report that human infants perceive certain vowels asymmetrically when tested with a repeating-background category-change operant headturn procedure. Specifically, discrimination is easier when the background is a more central vowel (e.g. /y/) and the target is a more peripheral vowel (e.g. /u/), compared to the opposite condition. We are testing human (adult native English listeners) and monkey discrimination of the German vowel pair /dut-dyt/, using both synthetic and natural stimuli (obtained from Polka and Bohn), and a within-subject design. So far, no significant asymmetries have emerged in the data using a percent correct measure. We will also report preliminary data from gerbils being trained in a between-subject design more comparable to that used with human infants. Plans are to continue testing all species until they reach asymptotic performance levels in order to determine if an RT measure will indicate vowel asymmetries. A video will be shown of humans, monkeys and gerbils working on the experiments.

Is the area measure a historical anomaly?

Green's well-known area theorem establishes an equivalence between the area under the yes-no ROC curve and the percent correct of an unbiased observer in a two-alternative forced-choice (2AFC) task with equivalent stimuli. In this article, we show that this conversion from yes-no detection data to hypothetical performance in a 2AFC task is unnecessary: The same yes-no detection data that are used to compute the area statistic can always be used to compute the percent correct of an unbiased observer in the yes-no detection task itself. We also show that the ROC curve may not be the ideal graphical device for many investigators. A more natural representation of the difficulty of a discrimination task is obtained by plotting the distribution of the posterior betting odds under equal base rates, which can be estimated from their distributions under unequal base rates. Finally, unlike the area measure and other traditional detection theory statistics, both the yes-no percent correct measure and the odds distributions generalize in an obvious and direct way to classification paradigms with more than two responses (e.g., identification).

Foveal contrast thresholds exhibit spatial-frequency- and polarity-specific contour interactions

Traditionally, contour interaction has been investigated at the visual acuity limit using a Landolt C and flanking bars, performance being quantified in terms of a percent correct measure. More recently, it has been shown that the properties of the contour interaction are different when larger stimuli are used: Contour interaction is not polarity specific, and spatial frequency tuning for an unflanked C is broader. Here we quantify contour interaction for stimuli 5x larger than the resolution limit in terms of contrast thresholds. We show that polarity of bars has little effect on unfiltered stimuli but does show very different effects on the spatial-frequency-tuning curves for discrimination of the Landolt C. This explains the polarity dependence of crowding at the visual acuity limit and its independence for larger unfiltered targets. Thus the underlying filtering function is composed of more than one mechanism, affected differently depending on the relative polarity of the test and flank contours.

Stochastic resonance in psychophysics and in animal behavior.

A recent analysis of the energy detector model in sensory psychophysics concluded that stochastic resonance does not occur in a measure of signal detectability ( d'), but can occur in a percent-correct measure of performance as an epiphenomenon of nonoptimal criterion placement [Tougaard (2000) Biol Cybern 83: 471-480]. When generalized to signal detection in sensory systems in general, this conclusion is a serious challenge to the idea that stochastic resonance could play a significant role in sensory processing in humans and other animals. It also seems to be inconsistent with recent demonstrations of stochastic resonance in sensory systems of both nonhuman animals and humans using measures of system performance such as signal-to-noise ratio of power spectral densities and percent-correct detections in a two-interval forced-choice paradigm, both closely related to d'. In this paper we address this apparent dilemma by discussing several models of how stochastic resonance can arise in signal detection systems, including especially those that implement a "soft threshold" at the input transform stage. One example involves redefining d' for energy increments in terms of parameters of the spike-count distribution of FitzHugh-Nagumo neurons. Another involves a Poisson spike generator that receives an exponentially transformed noisy periodic signal. In this case it can be shown that the signal-to-noise ratio of the power spectral density at the signal frequency, which exhibits stochastic resonance, is proportional to d'. Finally, a variant of d' is shown to exhibit stochastic resonance when calculated directly from the distributions of power spectral densities at the signal frequency resulting from transformation of noise alone and a noisy signal by a sufficiently steep nonlinear response function. All of these examples, and others from the literature, imply that stochastic resonance is more than an epiphenomenon, although significant limitations to the extent to which adding noise can aid detection do exist.

Impairment of long-term memory and sparing of short-term memory in monkeys with medial temporal lobe lesions: a response to Ringo

During the last decade, an animal model of human amnesia was developed in the monkey. Studies using this model have identified structures in the medial temporal lobe that are essential for forming long-term memory (i.e. the hippocampus and the entorhinal, perirhinal and parahippocampal cortices). Recently, an important aspect of these studies was questioned by Ringo (Behav. Brain Res., 42 (1991) 123-134). He suggested that the data from the delayed non-matching-to-sample task, which has been extensively used in these studies, have been analyzed in a potentially misleading way. He reanalyzed the data from several laboratories by transforming percent correct data to a discriminability (d') measure based on signal detection theory. In monkeys with lesions, performance appeared to be equivalently impaired at short and long retention delays. He concluded that the data do not support the idea that medial temporal lobe damage produces an impairment in long-term memory, but not short-term memory. However, most of the studies he analyzed were not designed to address the distinction between short-term and long-term memory. We show here that, in studies designed to compare short-term and long-term memory directly, medial temporal lobe lesions impair long-term memory while leaving short-term memory intact. This result is obtained whether the data are analyzed using a percent correct measure, the d' measure, or an arcsine transform.

A fully automated response time test of speech recognition

A fully automated computer-controlled word recognition test was developed to investigate both speech recognition and central processing capacity (ability to focus versus divide attention appropriately). The test consists of word lists that can be presented either diotically in the presence of a noise masker−for investigation of peripheral function; or dichotically−for investigation of central auditory function. Subjects monitor the word lists for words in prespecified categories. The percentage of correctly identified words is noted and the time taken to respond is measured and stored automatically on computer for later analysis. Three advantages of this test over conventional measures are: (1) its flexibility for use as a peripheral and/or a central test; (2) the measurement of response time, which appears to be more sensitive than the percent correct measure; and (3) the need for the subject to understand as well as to hear the test stimuli. [Work supported by a grant from the Veterans Administration.]

Lateralized Brain Dysfunction in Depression: Analysis of Memory

This study was designed to investigate the hypothesized association between non-dominant hemispheric dysfunction and depressive illness. Twenty-eight depressed inpatients, fourteen nonpsychiatric controls, and one right temporal lesioned control were administered both a verbal and visual list learning task with restrictive reminding. Each test consisted of four recall trials (employing a percent correct measure) and two recognition probes (employing a d' measure). Items to be remembered were names of animals on the verbal task and moderately complex designs on the visual task. Standardized difference scores (verbal-visual) were determined to control for motivational confounds and individual performance levels.

Sensitivity to sucrose in smokers and nonsmokers: A comparison of TSD and percent correct measures

Ten smokers and 10 nonsmokers were compared for taste sensitivity to sucrose on two discrimination tasks, one at near-threshold (NT) concentration level and one at suprathreshold (ST) concentration level. The data were analyzed using the ds and area under ROC curve measures based on the theory of signal detection (TSD). and by using the percentage correct responses. Analysis of both TSD measures showed smokers to be significantly more sensitive than nonsmokers on the NT .task and more sensitive on the ST task, but not significantly so. Comparison of mean percent correct responses showed no significant differences. A reflexive enhancement phenomenon was suggested as a possible interpretation. A decreased sensitivity to the bitter taste in smokers apparently results in an increased sensitivity to the sweet taste.