The undershoot bias: learning to act optimally under uncertainty

Learning in stochastic environments is increasingly viewed as an important psychological ability. To extend these results from a perceptual to a motor domain, we tested whether participants could learn to solve a stochastic minimal-time task using exploratory learning. The task involved moving a cursor on a computer screen to a target. We systematically varied the degree of random error in movement in three different conditions; each condition had a distinct time-optimal solution. We found that participants approximated the optimal solutions with practice. The results show that adults are sensitive to the stochastic structure of a task and naturally adjust the magnitude of an undershoot bias to the particular movement error of a task.

Analysis of probabilistic classification learning in patients with Parkinson's disease before and after pallidotomy surgery

This study examined the characteristics of probabilistic classification learning, a form of implicit learning previously shown to be impaired in patients with basal ganglia dysfunction (e.g., Parkinson's disease). In this task, subjects learn to predict the weather using associations that are formed gradually across many trials, because of the probabilistic nature of the cue-outcome relationships. Patients with Parkinson's disease, both before and after pallidotomy, and age-matched control subjects, exhibited evidence of probabilistic classification learning across 100 training trials. However, pallidotomy appears to hinder the learning of associations most implicit in nature (i.e., weakly associated cues). Although subjects were most sensitive to single-cue associations when learning the task, there is evidence that cue combinations contribute significantly to probability learning. The utility of multiple dependent measures is discussed.

Children are cursed: an asymmetric bias in mental-state attribution

Young children have problems reasoning about false beliefs. We suggest that this is at least partially the result of the same curse of knowledge that has been observed in adults--a tendency to be biased by one's own knowledge when assessing the knowledge of a more naive person. We tested 3- to 5-year-old children in a knowledge-attribution task and found that young children exhibited a curse-of-knowledge bias to a greater extent than older children, a finding that is consistent with their greater difficulty with false-belief tasks. We also found that children's misattributions were asymmetric. They were limited to cases in which the children were more knowledgeable than the other person; misattributions did not occur when the children were more ignorant than the other person. This suggests that their difficulty is better characterized by the curse of knowledge than by more general egocentrism or rationality accounts.

Partition priming in judgment under uncertainty

We show that likelihood judgments are biased toward an ignorance-prior probability that assigns equal credence to each mutually exclusive event considered by the judge. The value of the ignorance prior depends crucially on how the set of possibilities (i.e., the state space) is subjectively partitioned by the judge. For instance, asking "what is the probability that Sunday will be hotter than any other day next week?" facilitates a two-fold case partition, [Sunday hotter, Sunday not hotter], thus priming an ignorance prior of 1/2. In contrast, asking "what is the probability that the hottest day of the week will be Sunday?" facilitates a seven-fold class partition, [Sunday hottest, Monday hottest, etc.], priming an ignorance prior of 1/7. In four studies, we observed systematic partition dependence: Judgments made by participants presented with either case or class formulations of the same query were biased toward the corresponding ignorance prior.

Representing spatial location and layout from sparse kinesthetic contacts

Participants' fingers were guided to 2 locations on a table for 3 s, then back to the start. They reported distances and angles between the locations by (a) replacing 1 or 2 fingers, (b) translating the contacted configuration, or (c) estimating distance or angle alone. Distance error increased across these conditions. Angular error increased when the angular reference axis was rotated before the response. Replacing 1 finger was impaired by a change in posture from exposure to test. The results suggest a kinesthetic representation is used to replace the fingers, but to estimate distance and angle at new locations, a configural representation is computed. This presentation is oriented within an extrinsic reference frame and maintains shape more accurately than scale.

The multifaceted nature of unsupervised category learning

A substantial portion of category-learning research has focused on one learning mode--namely, classification learning (a supervised learning mode). Subsequently, theories of category learning have focused on how the abstract structure of categories (i.e., the co-occurrence patterns of feature values) affects acquisition. Recent work in supervised learning has shown that a learner's interactions with the stimulus set also plays an important role in acquisition. The present study extends this work to unsupervised learning situations involving simple one-dimensional stimuli. The results suggest that categorization performance is a function of both learning mode (i.e., study conditions) and learning problem (i.e., category structure). Unsupervised learning, like supervised learning, appears to be multifaceted, with different learning modes best paired with certain learning problems.

Syllogistic reasoning time: disconfirmation disconfirmed

Models of deductive reasoning typically assume that reasoners dedicate more logical analysis to unbelievable conclusions than to believable ones (e.g., Evans, Newstead, Allen, & Pollard, 1994; Newstead, Pollard, Evans, & Allen, 1992). When the conclusion is believable, reasoners are assumed to accept it without much further thought, but when it is unbelievable, they are assumed to analyze the conclusion, presumably in an attempt to disconfirm it. This disconfirmation hypothesis leads to two predictions, which were tested in the present experiment: Reasoners should take longer to reason about problems leading to unbelievable conclusions, and reasoners should consider more models or representations of premise information for unbelievable conclusions than for believable ones. Neither prediction was supported by our data. Indeed, we observed that reasoners took significantly longer to reason about believable conclusions than about unbelievable ones and generated the same number of representations regardless of the believability of the premises. We propose a model, based on a modified version of verbal reasoning theory (Polk & Newell, 1995), that does not depend on the disconfirmation assumption.

MDFT account of decision making under time pressure

In this paper, decision making under time pressure for multiattribute choice alternatives in a risky environment is investigated. A model, multiattribute decision field theory (MDFT), is introduced that describes both the dynamic and the stochastic nature of decision making and accounts for the observed changes in choice probabilities, including preference reversals as a function of time limit. An experiment in which five different time limits were imposed on the decision maker is presented to test the predictions of the model. It is shown that MDFT is able to account for the complex decision behavior observed in the data. Furthermore, MDFT is compared with the predictions of decision field theory (Busemeyer & Townsend, 1993; Roe, Busemeyer, & Townsend, 2001).

Judging confidence influences decision processing in comparative judgments

Current theories of confidence in human judgment assume that confidence and the decision it is based on are inextricably tied to the same process (decisional locus theories) or that confidence processing begins only once the primary decision has been completed (postdecisional locus theories). In the absence of auxiliary assumptions, however, neither class of theory permits the judgment of confidence to affect primary decision processing. In the present study, we examined the effect of rendering confidence judgments on the properties of the decision process in a sensory discrimination task. An examination of the properties of the time taken to determine confidence (i.e., the time taken to render the judgment of confidence) revealed clear evidence of postdecisional confidence processing. Concomitantly, the requirement of confidence judgments was found to substantially increase decisional response times, suggesting that some confidence processing occurs during the primary decision process. We discuss the implications of these findings for contemporary models of confidence in human judgment.

Conditionals and conditional probability

The authors report 3 experiments in which participants were invited to judge the probability of statements of the form if p then q given frequency information about the cases pq, p not q, not pq, and not p not q (where not = not). Three hypotheses were compared: (a) that people equate the probability with that of the material conditional, 1 - P(p not q); (b) that people assign the conditional probability, P(q/p); and (c) that people assign the conjunctive probability P(pq). The experimental evidence allowed rejection of the 1st hypothesis but provided some support for the 2nd and 3rd hypotheses. Individual difference analyses showed that half of the participants used conditional probability and that most of the remaining participants used conjunctive probability as the basis of their judgments.

A theoretical framework for understanding the effects of simultaneous base-rate and payoff manipulations on decision criterion learning in perceptual categorization

Observers completed perceptual categorization tasks in which base rates and payoffs were manipulated separately or simultaneously across a range of category discriminabilities. Decision criterion estimates from the simultaneous base-rate/payoff conditions were closer to optimal than those predicted from the independence assumption, in line with predictions from the flat-maxima hypothesis. A hybrid model that instantiated the flat-maxima and competition between reward and accuracy maximization hypotheses was applied to the data as well as used in a reanalysis of C. J. Bohil and W.J. Maddox's (2001) study. The hybrid model was superior to a model that incorporated the independence assumption, suggesting that violations of the independence assumption are to be expected and are well captured by the flat-maxima hypothesis, without requiring any additional assumptions.

Relationship between priming and recognition in deterministic and probabilistic sequence learning

Exposure to a repeating sequence of target stimuli in a speeded localization task can support both priming of sequence-consistent responses and recognition of sequence components. In 3 experiments with both deterministic and probabilistic sequences, the authors used a novel procedure in which measures or priming and recognition were taken concurrently and asked whether these measures can be dissociated. In all of these experiments, both measures were above chance at the group level and no evidence of dissociation was found. Item-level analyses of the data in Experiment 3 did reveal dissociations in that (a) recognition judgments were affected by response speed independently of old-new status and (b) items that were not discriminated in recognition nonetheless showed priming. However, the authors show that these data, together with the group-level results, are compatible with a formal model in which priming and recognition are based on a single common memory variable.

[Language acquisition and statistical learning]

Statistical learning is a basic mechanism of information processing in the human brain. The purpose lies in the extraction of probabilistic regularities from the multitude of sensory inputs. Principles of statistical learning contribute significantly to language acquisition and presumably also to language recovery following stroke. The empirical database presented in this manuscript demonstrates that the process of word segmentation, acquisition of a lexicon, and acquisition of simple grammatical rules can be entirely explained through statistical learning. Statistical learning is mediated by changes in synaptic weights in neuronal networks. The concept therefore stands at the transition to molecular biology and pharmacology of the neuronal synapse. It still remains to be shown if all aspects of language acquisition can be explained through statistical learning and which regions of the brain are involved in or capable of statistical learning. Principles of effective language training are obvious already. Most important is the massive, repeated interactive exposure. Conscious processing of the stimulus material may not be essential. The crucial principle is a high cooccurrence of language and corresponding sensory processes. This requires a more intense training frequency than traditional aphasia treatment programs provide.

Decision making in adolescents with behavior disorders and adults with substance abuse

OBJECTIVE

The study assessed the validity of the Gambling Task as a test of decision-making ability in adolescents and examined whether adolescents with behavior disorders, who are at risk for substance abuse, have deficits in decision making similar to those exhibited by adults with substance abuse.

METHOD

Performance on the Gambling Task in two testing sessions separated by 1 week was assessed in 64 12-14-year-old adolescents (31 healthy, 33 with externalizing behavior disorders) and 52 adults (22 healthy, 30 with substance abuse).

RESULTS

The healthy adolescents and the healthy adults had similar performance on the Gambling Task. Adolescents with behavior disorders performed more poorly than healthy adolescents, but only in the second testing session. In adults, overall Gambling Task performance did not differ between the healthy and substance abuse groups at either testing session, indicating no difference in learning of decision-making strategies between groups. However, adults with substance abuse performed more poorly than healthy adults during an early stage of the task, when participants presumably begin to understand the rewards and penalties involved in the task but are not yet sure of the actual risk of incurring penalities.

CONCLUSIONS

The Gambling Task can be used with adolescents. Testing with the Gambling Task revealed a deficit in decision making in adolescents with behavior disorders, who are at risk for substance abuse. This deficit may represent a vulnerability factor for the development of substance abuse.

Story understanding and counterfactual reasoning

D. Kahneman and A. Tversky (1982), in a seminal study on counterfactual reasoning, claimed empirical support for a simulation heuristic wherein ease of converting unusual conditions determines their selection as causes over normal conditions. Discourse analysis of their stories revealed a confounding of explanation and normality. A connectionist simulation of online comprehension and memory access of alternative conditions without conversion accounted for their data. Normality and explanation were varied independently in 2 experiments. Explanation but not normality affected the rank ordering of counterfactual conditions after reading. Access of alternative conditions in simulation was again the best predictor of empirical findings. Comprehension and memory operate where stories communicate information for decision making such as counterfactual reasoning and hindsight bias.

Myopic Social Prediction and the Solo Comparison Effect

Four experiments explored the psychological processes by which people make comparative social judgments. Each participant chose how much money to wager on beating an opponent on either a difficult or a simple trivia quiz. Quiz difficulty did not influence the average person's probability of winning, yet participants bet more on a simple quiz than on a difficult quiz in the first 3 experiments. The results suggest that this effect results from a tendency to attend more closely to a focal actor than to others. Experiment 4 directly manipulated focusing; when participants were led to focus on the opponent instead of themselves, the effect was reversed. The discussion relates the results to other literatures including overly optimistic self-evaluation, false consensus, overconfidence, and social comparison.

Developmental dyslexics show altered allocation of attention in visual classification tasks

OBJECTIVES

Event-related brain potentials (ERPs) were recorded to investigate allocation of attention in adult developmental dyslexics.

SUBJECTS AND METHODS

Twelve adult developmental dyslexics and 12 matched normal controls performed three visual choice reaction tasks. In the passive oddball condition, subjects watched two different simple visual stimuli presented with 87.5 and 12.5% probability. In the active oddball condition, participants responded to the low-probability target stimulus. In the active 50/50-condition, both stimuli were presented with 50% probability and a response was required to the target stimulus only.

RESULTS

No group differences emerged for performance, P300 latency or laterality and for N200 amplitude, latency or laterality. An enhancement of P300 amplitude with a frontal distribution was found for NoGo (standard)-stimuli in both active conditions for the dyslexic sample.

CONCLUSION

Results are discussed in the context of deviances in allocation of attentional resources in dyslexic readers.

A Hybrid-Similarity Exemplar Model for Predicting Distinctiveness Effects in Perceptual Old-New Recognition

In 2 sets of experiments, the authors investigated the basis for old-item distinctiveness effects in perceptual recognition, whereby distinctive old items are recognized with higher probability than are typical old items. In Experiment 1, distinctive old items were defined as those lying in isolated regions of a continuous-dimension similarity space. In this case, any beneficial effects of distinctiveness were absent or small, regardless of the structure of the test list used to assess recognition memory. In Experiment 2, distinctive items were defined as those objects containing certain discrete, individuating features. In this case, large old-item distinctive effects were observed, with the nature of the effects being modulated by the structure of the test lists. A hybrid-similarity exemplar model, combining elements of continuous-dimension distance and discrete-feature matching, was used to account for these distinctiveness effects in the recognition data.

Evaluation of six multinomial models of conscious and unconscious processes with the recall-recognition paradigm

Six multinomial processing-tree models (W. H. Batchelder & D. M. Riefer, 1999), which include parameters representing conscious and unconscious memory processes, were tested using the recall-recognition paradigm. Data from 2 experiments were fit equally well by 3 of the 6 models. One model recognition was an extension of the generate-recognize model (L. L. Jacoby, 1998), and another was an extension of the non-high-threshold model (D. M. McBride & B. A. Dosher, 1999). The 3rd model was the source evaluation model (D. M. McBride & B. A. Dosher, 1999). Values of the parameters of 2 of these 3 models, excepting the non-high-threshold model, responded to experimental manipulations in accordance with the meaning of the parameters. The equivalence of models with regard to goodness-of-fit tests is discussed as is how experiments can be designed to demonstrate the superiority of one model over another. The potential usefulness of these models in the study of amnesia is considered.

Intermanual transfer of procedural learning after extended practice of probabilistic sequences

Previous studies using simple, repeating patterns have suggested that the knowledge gained in early sequence learning is not effector-specific in that it transfers to muscle groups other than those used during training. The current experiments extended these findings to transfer after extensive practice with probabilistic sequences using a task on which people fail to gain declarative knowledge of the regularity. Specifically, an alternating serial reaction time (ASRT) task was used in which predictable and unpredictable trials alternated. Participants responded for the first five sessions using their right hand, then switched to the left hand for the sixth session. Stimuli were spatial in the first experiment and nonspatial in the second. Significant near-perfect transfer of pattern knowledge was seen in both experiments, suggesting that muscle-specific information for either the fingers or the eyes cannot explain the observed learning.

A Causal-Model Theory of Conceptual Representation and Categorization

This article presents a theory of categorization that accounts for the effects of causal knowledge that relates the features of categories. According to causal-model theory, people explicitly represent the probabilistic causal mechanisms that link category features and classify objects by evaluating whether they were likely to have been generated by those mechanisms. In 3 experiments, participants were taught causal knowledge that related the features of a novel category. Causal-model theory provided a good quantitative account of the effect of this knowledge on the importance of both individual features and interfeature correlations to classification. By enabling precise model fits and interpretable parameter estimates, causal-model theory helps place the theory-based approach to conceptual representation on equal footing with the well-known similarity-based approaches.

The effect of stimulus probability on pupillary response as an indicator of cognitive processing in human learning and categorization

Stimuli presented with a low probability of occurrence elicit larger pupillary dilations than those presented with a high frequency. Utilizing this stimulus probability effect, we conducted two Go/NoGo reaction time experiments to analyze category learning and compound coding in the context of associative learning. Experiment 1 showed no stimulus probability effect when participants used an abstract rule to classify stimuli with different probabilities into the same category. When no rules could be applied, however, the stimulus probability effect was observed. Hence, this effect can be utilized to identify the application of rules in category learning. Experiment 2 demonstrated that the stimulus probability effect can also be used in determining whether compounds are processed in terms of elemental stimuli or as new entities. The results of Experiment 2 supported elemental rather than configural processing. Obviously, the probability effect can be used as a tool for the evaluation of categorization and compound processing.

When wrong predictions provide more support than right ones

Correct predictions of rare events are normatively more supportive of a theory or hypothesis than correct predictions of common ones. In other words, correct bold predictions provide more support than do correct timid predictions. Are lay hypothesis testers sensitive to the boldness of predictions? Results reported here show that participants were very sensitive to boldness, often finding incorrect bold predictions more supportive than correct timid ones. Participants were willing to tolerate inaccurate predictions only when predictions were bold. This finding was demonstrated in the context of competing forecasters and in the context of competing scientific theories. The results support recent views of human inference that postulate that lay hypothesis testers are sensitive to the rarity of data. Furthermore, a normative (Bayesian) account can explain the present results and provides an alternative interpretation of similar results that have been explained using a purely descriptive model.

Influence of stimulus distance in implicit learning of spatial and nonspatial event sequences

The serial reaction time task has been widely used to investigate implicit learning mechanisms. In the present study, we investigated the effect of stimulus distance on learning of a spatial sequence independent of a sequence of responses. Participants had to respond to objects appearing at four different locations. The objects were presented in a sequence of nine elements, whereas the location at which an object was presented followed a sequence of eight elements. Thus, the spatial and the object sequences were independent of each other. Four groups of subjects for whom the distances of the locations chosen to present objects on the computer screen (3 cm, 6 cm, 12 cm, or 22 cm) differed were tested. Only the nonspatial sequence was learned as indicated by enhanced response latencies in nonsequenced random blocks. Stimulus distance had no effect on the amount of sequence learning. Additional analyses for subgroups of subjects who did not show explicit knowledge of the sequences after completion of the task indicated that for implicit learners also, sequence learning was not influenced by stimulus distance. The results are discussed with respect to current theories of implicit serial learning.

On the impact of different number representations in the number bisection task

The number bisection task--in which the numerical middle of two other numbers has to be determined (e.g., 6 is the numerical middle of 4 and 8)--has recently been used in neuropsychological case studies in addition to the number comparison task to assess quantitative capabilities in number processing. While factors determining difficulty in normal participants have often been systematically investigated in the number comparison task, this has not yet been done in the number bisection task. Based on a pilot study by van Herten (1999), we extracted four factors that we hypothesized to modulate difficulty in a verification version of the number bisection task in normal participants: Multiplicativity (whether or not the three numbers are part of a multiplication table), range (distance between the smallest and the largest number) for bisectable number triplets as well as distance of the middle number to the numerical middle, and bisection possibility for non-bisectable triplets. In the current study, we obtained large effects for all four factors on RT and accuracy data and some interesting interactions in normal participants. In a regression analysis, we additionally observed inhibitory effects of decade crossing and of ten inclusion (whether or not one of the three numbers was a decade number). We discuss the impact of these results for the use of the number bisection task as an assessment tool in neuropsychology. Finally, we examine the consequences of these results for models of number processing.