Robust averaging during perceptual judgment

Abstract

An optimal agent will base judgments on the strength and reliability of decision-relevant evidence. However, previous investigations of the computational mechanisms of perceptual judgments have focused on integration of the evidence mean (i.e., strength), and overlooked the contribution of evidence variance (i.e., reliability). Here, using a multielement averaging task, we show that human observers process heterogeneous decision-relevant evidence more slowly and less accurately, even when signal strength, signal-to-noise ratio, category uncertainty, and low-level perceptual variability are controlled for. Moreover, observers tend to exclude or downweight extreme samples of perceptual evidence, as a statistician might exclude an outlying data point. These phenomena are captured by a probabilistic optimal model in which observers integrate the log odds of each choice option. Robust averaging may have evolved to mitigate the influence of untrustworthy evidence in perceptual judgments.decision making | diffusion model | information integration P erceptual judgments typically involve a deliberative process in which evidence concerning the current state of the external world is considered. Over recent years, the twin goals of characterizing the computational mechanisms and the neural representations underlying this deliberation have come to the fore (1, 2). Because sensory evidence coming from the external world is intrinsically noisy, decisions will benefit from repeated sampling and accumulation of the collected evidence (3-5). Mathematical modeling studies support the view that serial sampling is a basic principle of choice behavior (3, 6-9), and recent neurophysiological recordings identify the parietal cortex as a candidate site for evidence accumulation in psychophysical tasks (10-14). However, the precise computations by which a decision variable (DV) is constructed and updated during decision making remain controversial (2, 4).One popular framework posits that integration is a simple summation process under which choices and their latencies depend linearly on the strength of sensory input (3,15,16). This mechanism is often illustrated by analogy to a court of law, where the jury tots up evidence for or against a guilty verdict (1). However, in a stochastic environment, committing to an action on the basis of evidence strength alone can be suboptimal, because evidence may strongly favor one option over another just by chance (17, 18). Rather, a statistically optimal policy is to base decisions on independent estimates of the strength (i.e., mean) and reliability (i.e., variance) of the currently extant sensory evidence, just as a researcher might compare two samples of data on the basis of an inferential statistic rather than merely calculating their central tendencies (19). To continue the courtroom analogy, a shrewd jury will consider not only how incriminating evidence is, but also the trustworthiness of the source of the evidence. These two factors are not necessarily correlated: For example, severely indic...

Priming by the variability of visual information

“…There was no prime × target interaction on accuracy [F (1,39) = 0.775, P = 0.384], suggesting that this effect is a facilitation in processing, rather than a because of a change in speed-accuracy trade-off. The variance of the prime did not impair accuracy, whereas the target variance did [F (1,39) = 16.0, P < 0.001], as previously reported (18).…”

“…2B), and t tests were used to assess the deviance of the resulting parameter estimates from zero. Parameter estimates associated with prime array variance and target array variance were both positive [prime: t (39) = 4.24, P < 0.001; target: t (39) = 6.54, P < 0.001], consistent with the previously described detrimental impact of high-variance arrays on decision latencies (18), whereas those associated with interaction between prime and target variance were negative [t (39) = 4.79, P < 0.001], consistent with the abovementioned observation that similar variance in the prime and the target facilitated responding. Crucially, these effects persisted even when the eight element-specific differences had been partialled out, indicating that it is a summary statistical representation-not feature information-that is driving priming by the variance.…”

“…To date, a great deal of research has focused on demonstrating that observers encode the central tendency of sensory data: humans can accurately estimate mean feature information from arrays of simple features (11,13,(18)(19)(20)(21), complex shapes or letters (22), or human faces (23,24), even when discrimination of individual array elements is at chance. However, limited consideration has been given to whether the visual system represents the dispersion (variability) of available visual information.…”

Salience driven value integration explains decision biases and preference reversal

“…To mathematically capture this pattern, we assume that the values, V i , for each alternative, i, are weighed by their momentary ranks and integrated in separate leaky accumulators with preference states P i (t) (extending Eq. 1): P i ðtÞ ¼ λ P i ðt − 1Þ þ ½V i ðtÞ · wðrank i ðtÞÞ þ Nð0; σÞ; [2] with w(max) > 1 and w(min) = 1 in selection and w(max) < 1 and w(min) = 1 in rejection decisions; rank i (t) is the momentary rank of item i at time t. In selection decisions, the alternative associated to the accumulator with the highest P i is chosen, whereas in rejection, the accumulator with the lowest P i is eliminated. This model accounts for the data in experiments 2 and 3 (purple circles in Figs.…”

“…We show that many known choice anomalies may arise from the microstructure of the value integration process. decision making | decoy effects | value psychophysics | expanded judgement R ecent research on the psychology and neuroscience of simple, evidence-based choices (e.g., integrating perceptual or reward information) has made impressive progress, leading to the conclusion that the brain is optimized to make the fastest decision for a specified accuracy (1)(2)(3)(4)(5). Accordingly, the observer is assumed to infer the most probable cause of a perceived experience by sequentially accumulating samples of noisy evidence until a response criterion is reached.…”