Learning From Peers’ Eye Movements in the Absence of Expert Guidance: A Proof of Concept Using Laboratory Stock Trading,Eye Tracking,and Machine Learning

Existing research shows that people can improve their decision skills by learning what experts paid attention to when faced with the same problem. However, in domains like financial education, effective instruction requires frequent, personalized feedback given at the point of decision, which makes it time‐consuming for experts to provide and thus, prohibitively costly. We address this by demonstrating an automated feedback mechanism that allows amateur decision‐makers to learn what information to attend to from one another, rather than from an expert. In the first experiment, eye movements of N = 100 subjects were recorded while they repeatedly performed a standard behavioral finance investment task. Consistent with previous studies, we found that a significant proportion of subjects were affected by decision bias. In the second experiment, a different group of N = 100 subjects faced the same task but, after each choice, they received individual, machine learning‐generated feedback on whether their pre‐decision eye movements resembled those made by Experiment 1 subjects prior to good decisions. As a result, Experiment 2 subjects learned to analyze information similarly to their successful peers, which in turn reduced their decision bias. Furthermore, subjects with low Cognitive Reflection Test scores gained more from the proposed form of process feedback than from standard behavioral feedback based on decision outcomes.     

Control of children's observing responses by information feedback during visual discrimination learning

Four experiments examined the control of observing responses by information feedback during visual discrimination learning. Second-grade children participated in Experiment 1; kindergarten, second-, and fifth-grade children were subjects in Experiments 2 and 3, and grade 5 and adult subjects were tested in Experiment 4. In order to view the stimuli, subjects in Experiments 1, 2, and 3 activated lights in viewing boxes; in Experiment 4, stimulus fixations were measured using a corneal reflection technique. Fifth graders and adults observed the discriminative stimuli for longer times on trials following negative feedback than on trials following positive feedback; in contrast, kindergartener's observing was not affected by type of feedback. Second graders showed smaller and less reliable reactions to type of feedback than did older subjects. These results support the view that visual observing is controlled by cognitive processes associated with hypothesis testing.     

Causal attribution, cognitive interference, and the generalization of learned helplessness

We assessed the role of off-task cognitions in mediating the performance effects of global and specific attributions for failure. In Experiment 1, subjects were divided into global and specific attributors and were exposed to either no feedback or failure feedback. In Experiment 2, subjects were exposed to no feedback or to unsolvable problems wherein they received attribution for failure to specific or global causes. Experiment 3 added a condition in which subjects were restrained from engaging in off-task cognitions. Results showed that exposure to unsolvable problems deteriorated performance and increased off-task cognitions mainly among subjects who attributed failure to global causes. In addition, the enhancement of off-task cognitions interfered with performance following unsolvable problems. The introduction of instructions that discouraged subjects from engaging in off-task cognitions eliminated the detrimental effects of global attribution. Results are discussed in terms of test anxiety and excuse-making conceptualizations of learned helplessness.     

The generation effect: Further tests of the lexical activation hypothesis

Three experiments compared recognition memory for word versus nonword responses when they had been either read or generated using a rhyme rule and either a word or nonword stimulus. That is, either the wordshop or the nonwordthop was generated from either the wordchop or the nonwordphop. In Experiment 1, the lexicality of the stimulus and response terms was manipulated orthogonally between subjects; in Experiments 2 and 3, it was manipulated orthogonally within subjects. In Experiment 3, subjects also made a lexical (word-nonword) decision about each response term after it had been read or generated. In all three experiments, memory performance was better for generated than read responses. This generation effect occurred only if the response term was a word, regardless of whether the stimulus term was a word or a nonword. These results are discussed in terms of the roles that lexical memory and response unitization play in the generation effect.     

"Eating lightly" and the self-presentation of femininity

In Experiment 1, male and female subjects were given an opportunity to snack as they participated in a "get-acquainted study" with a same-sex or opposite-sex partner (confederate) whose social desirability was manipulated. Consistent with the hypothesis that women may eat less when motivated to present themselves in a feminine light, female subjects ate significantly less with a desirable male partner than in the remaining three conditions. In contrast, male subjects did not eat more (or less) with a desirable woman, although they did show an overall tendency to eat less with female (vs. male) partners. In Experiment 2, female subjects snacked as they got acquainted with a desirable male partner (confederate). Before this interaction, subjects received feedback indicating that they had either very masculine or very feminine interests. In addition, subjects believed either that their male partner was aware of their gender feedback or that he was unaware. Consistent with predictions derived from Schlenker's (1982) analytic-identity theory of social conduct, subjects in the partner-aware conditions ate less when they had received masculine (vs. feminine) feedback, whereas subjects in the partner-unaware conditions ate less when they had received feminine (vs. masculine) feedback. Implications for understanding eating disorders such as anorexia and bulimia are discussed.     

Temporal effects on attributions for one's own behavior: The role of task outcome

Previous investigations into how attributions for one's own behavior change over time have resulted in surprisingly inconsistent results. Two experiments were conducted to account for these discrepant findings. In Experiment 1 male undergraduates were given feedback indicating that they had done either well or poorly on a skill-assessment test. Half of the subjects believed they were being videotaped when performing the test, half did not. In addition, half of the subjects completed attribution questionnaires immediately after the feedback, whereas half completed the questionnaires 2 or 3 days later. It was found that subjects who felt they had succeeded on the task made attributions that were more dispositional over time and subjects who felt they had failed made attributions that were more situational over time. No effect for the videotape manipulation was found. Experiment 2 replicated the task outcome effect and provided evidence suggesting that the effect was caused by a selective forgetting of unflattering attributions.     

Processing a display even after you make a response to it. How perceptual errors can be corrected

We know that when people make responses which they did not intend they can discover this by monitoring kinaesthetic and visual feedback. It is less clear whether they can also correct perceptual errors which occur when they mistake one signal for another. It was argued that, if they can sometimes do this, extra errors which occur when discriminations become more difficult may be detected and corrected. Experiment I compared the ability of young fit subjects to detect errors made during easy and during difficult discriminations between tone signals. There was no evidence that any additional errors made to difficult discriminations were detected. Fast errors were detected, slower errors were not. The results were consistent with the idea that subjects can detect fast motor errors by monitoring feedback, but that they cannot detect perceptual mistakes.

In Experiment II subjects made easy and difficult line length discriminations. Displays lasted for 100 ms, 200 ms or 500 ms and were followed by random masks. In this case, fast errors were again corrected more frequently than slow errors but eight aspects of the data suggest that subjects could, and did, correct perceptual as well as motor mistakes and that they managed to do this by continuing to process a display after the moment at which they may have initiated an impulsive response to it. The data are interpreted in terms of a “Committee Decision” model for extended perceptual processing previously applied to other, similar data by Rabbitt and Rodgers (1977) and Rabbitt, Cumming and Vyas (1978).     

Effects of interpolating a secondary task during the intertrial interval of a concept identification problem

Several tasks were interpolated during the intertrial interval (ITI) of a concept identification (CI) task. In Experiment I, 234 subjects were required to count backwards by threes, emit a sibilant, or to refrain from thinking after all responses, after correct responses, or after errors. Counting backwards interfered with performance, but only if it occurred after errors. Stimulus and feedback information was either withheld, as in Experiment I, or made available during the ITI in Experiment II. Results of the data from 280 subjects indicated that counting backwards interfered with performance only when interpolated after errors, when stimulus-feedback information was withheld, but had an equivalent effect after errors and correct responses when stimulus-feedback information was available during ITI, indicating that different strategies were induced by the presence of such information.     

The utilization of visual feedback information during rapid pointing movements

Three experiments were conducted to determine how variables other than movement time influence the speed of visual feedback utilization in a target-pointing task. In Experiment 1, subjects moved a stylus to a target 20 cm away with movement times of approximately 225 msec. Visual feedback was manipulated by leaving the room lights on over the whole course of the movement or extinguishing the lights upon movement initiation, while prior knowledge about feedback availability was manipulated by blocking or randomizing feedback. Subjects exhibited less radial error in the lights-on/blocked condition than in the other three conditions. In Experiment 2, when subjects were forced to use vision by a laterally displacing prism, it was found that they benefited from the presence of visual feedback regardless of feedback uncertainty even when moving very rapidly (e.g. less than 190 msec). In Experiment 3, subjects pointed with and without a prism over a wide variety of movement times. Subjects benefited from vision much earlier in the prism condition. Subjects seem able to use vision rapidly to modify aiming movements but may do so only when the visual information is predictably available and/or yields an error large enough to detect early enough to correct.     

Derived relational responding as generalized operant behavior

The major aim of the present study was to demonstrate that derived relational responding may be viewed as a form of generalized operant behavior. In Experiment 1, 4 subjects were divided into two conditions (2 in each condition). Using a two-comparison matching-to-sample procedure, all subjects were trained and tested for the formation of two combinatorially entailed relations. Subjects were trained and tested across multiple stimulus sets. Each set was composed of novel stimuli. Both Conditions 1 and 2 involved explicit performance-contingent feedback presented at the end of each block of test trials (i.e., delayed feedback). In Condition 1, feedback was accurate (consistent with the experimenter-designated relations) following exposure to the initial stimulus sets. When subjects' responding reached a predefined mastery criterion, the feedback then switched to inaccurate (not consistent with the experimenter-designated relations) until responding once again reached a predefined criterion. Condition 2 was similar to Condition 1, except that exposure to the initial stimulus sets was followed by inaccurate feedback and once the criterion was reached feedback switched to accurate. Once relational responding emerged and stabilized, response patterns on novel stimulus sets were controlled by the feedback delivered for previous stimulus sets. Experiment 2 replicated Experiment 1, except that during Conditions 3 and 4 four comparison stimuli were employed during training and testing. Experiment 3 was similar to Condition 1 of Experiment 1, except that after the mastery criterion was reached for class-consistent responding, feedback alternated from accurate to inaccurate across each successive stimulus set. Experiment 4 involved two types of feedback, one type following tests for mutual entailment and the other type following tests for combinatorial entailment. Results from this experiment demonstrated that mutual and combinatorial entailment may be controlled independently by accurate and inaccurate feedback. Overall, the data support the suggestion, made by relational frame theory, that derived relational responding is a form of generalized operant behavior.     

A Comparison of Written,Vocal, and Video Feedback When Training Teachers

We compared the effectiveness of and preference for different feedback strategies when training six special education teachers during a 5-day summer training program. In Experiment 1, teachers received written or vocal feedback while learning to implement two different types of preference assessments. In Experiment 2, we compared either written or vocal feedback to vocal feedback combined with video replays of performance while training the teachers to implement two variations of discrete-trial training. The different forms of feedback were similarly effective for all of the participants. However, all of the teachers showed a preference for vocal feedback when they could choose which form of feedback they would receive from the experimenter. Results have important implications for the use of feedback when training teachers.     

Awareness of the response after feedback training for changes in heart rate and sudomotor laterality

What is the relation between the ability to control visceral responding on a biofeedback task and the ability to report behaviors actually contributing to this performance? Subjects received biofeedback training for unidentified visceral responses and then gave written reports about what they had done to control the feedback displays. Independent judges were given these reports and, on the basis of knowledge about activities known to contribute to visceral activity, were asked to determine the visceral responses for which the subjects had been trained. The reports of subjects who succeeded at bidirectional control of heart rate (Experiment 1) or sudomotor laterality (Experiment 2) showed awareness of behaviors related to feedback as assessed by this procedure, whereas the reports of subjects who failed at bidirectional control did not. Subsequent experiments indicated that these results did not depend on a learning strategy that might have been specific to the initial studies. These findings call into question the view that people are unaware of what they have done to produce the response after training on biofeedback tasks. Earlier studies reporting lack of awareness in biofeedback are discussed in light of factors that affect the measurement of biofeedback learning and response awareness.     

“Level of Aspiration”: Preference for Time Versus Response Feedback and History Effects on the Setting of Personal Performance Standards

Human subjects responded on a computer keyboard and accumulated points according to fixed-ratio (FR) reinforcement schedules. In Experiment 1, subjects responded under a FR 500 schedule. Under baseline conditions satisfying the schedule requirement resulted in counter points and session termination. Subsequently, subjects could (a) choose to have a clock appear on the screen during the interreinforcement interval (IRI) as well as to enter a target time which they would attempt to better during that session, (b) choose to enter a target time or respond under baseline conditions, and (c) enter a target time and choose between having a clock appear throughout or at the end of experimental sessions. In Experiment 2, subjects responded under a FR 500 schedule, entered a target time each session, and could respond during the session to briefly produce either (a) clock feedback, or (b) the number of responses emitted by the subject. In Experiment 3, subjects responded under FR 250, 500, 1000, and 2000 schedule parameters, entered target times and responded for either clock or response feedback. Subjects (a) preferred responding under conditions in which target times were entered to responding under baseline conditions, (b) preferred to have the clock illuminated throughout rather than at the end of experimental sessions, (c) preferred response to clock feedback under all schedule parameters, (d) responded to having equaled or bettered a target time by lowering target time for the subsequent session, and (e) responded to having missed a target time by maintaining the same time during the subsequent session. The results were interpreted within the context of behavior analytic as opposed to more traditional personality theory.

     

The Role of Task Experience and Prior Knowledge for Detecting Invalid Augmented Feedback while Learning a Motor Skill

Two hypotheses were investigated. These were generated from results reported by Buekers, Magill, and Sneyers (1994) and an uncertainty account of those and other effects on skill learning of erroneous knowledge of results (KR). The first hypothesis proposes that if experienced performers have developed the capability of detecting and correcting errors, then they should not be influenced by erroneous KR in the same way as novices. The second proposes that if information about the invalid feedback is given to subjects prior to the beginning of the practice trials, then the capability for assessing feedback can be accelerated such that subjects who receive this advance information should not be misled by the erroneous KR. The first hypothesis was investigated in Experiment 1 by having subjects practise an anticipation timing task for 450 trials. One group received correct KR on all trials, and another group received no KR. Half of the subjects in each group were unexpectedly switched to erroneous KR after 400 trials. Results indicated that the erroneous KR influenced both groups during acquisition and on a retention test one minute later. However, on a retention test given one week later, only the group that had practised with no KR before being switched to erroneous KR continued to show the negative influence of the invalid KR. The second hypothesis was examined in Experiment 2 by telling subjects prior to beginning practice trials that it was possible that the KR they received would be erroneous. Results showed that this intervention strategy was effective on the no-KR retention tests only for subjects who had no-KR trials alternated with trials on which they received the erroneous KR. The results of these experiments provide evidence that uncertainty about the validity of sensory feedback increases the likelihood that subjects will be misled by invalid augmented feedback. Moreover, this uncertainty can be overcome by increased experience or by providing advance knowledge about the nature of the invalid feedback.     

Retrieving names in old age: Short- and (very) long-term effects of repetition

Two experiments are reported that examine the effects of repetition on name retrieval in younger adults (in their 50s and 60s) and older adults (in their 70s and 80s). In Experiment 1, the subjects were asked to name a set of famous faces four times over the course of a 1-h session. Younger subjects produced significantly more correct responses than did older subjects. There was significant improvement with repeated attempts at naming, with younger and older subjects benefiting equally in terms of increasing numbers of correct responses across the session. In contrast, there was a highly significant age deficit in picture recognition over a similar retention interval. A qualitative analysis of naming responses (full name vs. part of the name) provided support for the view that aging and nonrecent use have equivalent effects on retrieval. In Experiment 2, younger subjects (but not older subjects) were significantly more likely to correctly name famous faces that they had seen 22 months previously than to correctly name new famous faces. In contrast, older subjects (but not younger subjects) were significantly more likely to produce erroneous names to famous faces that they had seen 22 months previously than to new famous faces. It is concluded that repetition priming may be relatively unaffected by aging over short retention intervals (Experiment 1) but not over a very long retention interval (Experiment 2).     

MODIFYING LETTER AND NUMBER REVERSALS IN ELEMENTARY SCHOOL CHILDREN1

Children from regular and special-education classes were exposed to tutoring procedures designed to modify letter and number reversals. In Experiment I, two students showed reversals during letter-naming (e.g., saying p for q) and two-digit number (e.g., writing 31 for 13) exercises. In Experiment II, one subject made two-digit reversals, while the other child reversed single-digit numbers (e.g., 9 written with circle on right side of stem). The subject in Experiment III emitted letter reversals under naming, dictation, and copying exercises. The child in Experiment IV showed written letter reversals on both sides of the “body midline” under dictation procedures. Basic treatment procedures during all experiments involved modelling of correct and incorrect symbol formation, and differential experimenter feedback following student responses. During Experiments I and II, subjects were praised for each correct response, and informed when a reversal occurred. Experiment III included a Modelling Only Phase before these feedback procedures. Besides praise, treatment conditions in Experiment IV consisted of additional feedback following correct responses (e.g., charting correct responses), and temporal delays imposed between dictated letters and student responses. Except for occasional letter-naming reversals of one subject (Experiment I), and letter dictation errors of another (Experiment IV), reversals were eliminated when all experiments terminated. The Modelling Only Phase in Experiment III also reduced reversals to a low level. These findings suggest that a variety of reversal problems may be effectively assessed and remediated via simple modelling and reinforcement procedures. In addition, postcheck observations indicated that the effects were enduring. The present procedures should be easily implemented by school personnel.     

Knowing but not remembering: Adult age differences in recollective experience

Age differences in recollective experience were examined in two experiments in which younger and older adults used their self-generated associations as retrieval cues. When recalling an item, subjects indicated whether they consciously remembered its prior occurrence, or merely knew that it was presented previously. The results of both experiments showed that aging selectively impaired retention accompanied by recollective experience, as measured byremember responses, but had no effects in the absence of recollective experience, as measured byknow responses. In Experiment 2, a similar pattern of data was obtained for a group of younger adults by increasing the rate of presentation at study. The results also indicated that judgments of recollective experience were related to type of encoding: Subjects who generated detailed associations reported higher levels of remember responses and lower levels of know responses than did subjects who generated fewer detailed associations. The results are discussed in terms of processes related to perceptual familiarity and contextual detail.     

Can eyewitnesses correct for external influences on their lineup identifications? The actual/counterfactual assessment paradigm

Real-world eyewitnesses are often asked whether their lineup responses were affected by various external influences, but it is unknown whether they can accurately answer these types of questions. The witness-report-of-influence mental-correction model is proposed to explain witnesses' reports of influence. Two experiments used a new paradigm (the actual/counterfactual paradigm) to examine eyewitnesses' abilities to report accurately on the influence of lineup manipulations. Eyewitnesses were administered either confirming feedback or no feedback (Experiment 1, n = 103), or a cautionary instruction or no cautionary instruction (Experiment 2, n = 114). Eyewitnesses then gave actual responses (retrospective confidence, view, and attention measures in Experiment 1; identification decision in Experiment 2) as well as counterfactual responses stating how they would have responded in the alternative condition. Results across both studies showed an asymmetric estimation of influence pattern: Eyewitnesses who received an influencing manipulation estimated significantly less of a change in their responses than eyewitnesses who did not receive an influencing manipulation. A 48-hr delay between actual and counterfactual responses did not moderate any effects. Results are explained by witnesses' implicit theories of influence.     

Short-term memory processes in counting

Four experiments examined the memory processes used to maintain location in a counting sequence. In Experiment 1, subjects who rapidly counted forward omitted many repeated-digit numbers (e.g., 77), as found previously with backward counting. Subjects in Experiment 2 counted backward with normal auditory feedback or with headphones through which white noise was channeled. In both cases, repeated-digit errors predominated, suggesting that the contents of short-term memory, rather than auditory sensory memory, are checked during counting. In Experiment 3, subjects silently wrote counting responses, and the omission errors resembled those in vocal counting. Repetition errors were also found and attributed to phonological recoding failures. Articulatory suppression in Experiment 4 greatly increased the number of repetition errors in the written counting task. A model of the counting process was proposed according to which subjects keep track of their location in the counting sequence by monitoring phonologically coded short-term memory representations of the numbers.     

Demonstrations of a generation effect in context memory

Generation often leads to increased memorability within a laboratory context (see, e.g., Slamecka & Graf, 1978). Of interest in the present study is whether the benefits of generation extend beyond item memory to context memory. To investigate this question, in three experiments, we asked subjects to remember in which of two contexts they had read or generated words. In Experiment 1, the contexts were two different rooms; in Experiment 2A, the contexts were two different computer screens; in Experiment 2B, the contexts were different perceptual characteristics of the to-be-remembered words. In all experiments, subjects were better at remembering the context of generated words than of read words.