Greg Barron

On Adaptation, Maximization, and Reinforcement Learning Among Cognitive Strategies

Analysis of binary choice behavior in iterated tasks with immediate feedback reveals robust deviations from maximization that can be described as indications of 3 effects: (a) a payoff variability effect, in which high payoff variability seems to move choice behavior toward random choice; (b) underweighting of rare events, in which alternatives that yield the best payoffs most of the time are attractive even when they are associated with a lower expected return; and (c) loss aversion, in which alternatives that minimize the probability of losses can be more attractive than those that maximize expected payoffs. The results are closer to probability matching than to maximization. Best approximation is provided with a model of reinforcement learning among cognitive strategies (RELACS). This model captures the 3 deviations, the learning curves, and the effect of information on uncertainty avoidance. It outperforms other models in fitting the data and in predicting behavior in other experiments.

The Role of Personal Experience in Contributing to Different Patterns of Response to Rare Terrorist Attacks

An examination of the behavioral effect of repeated terrorist attacks reveals that local residents (of the attacked area) appear to be much less sensitive to this risk than international tourists. Furthermore, the limited sensitivity on the part of local residents seems to diminish with time, even when the attacks continue. An experimental study shows a similar pattern in a laboratory experiment that focuses on a basic decision task: when making a single decision based on a description of the problem, people tend to be more risk averse. Personal experience with the problem reduces this sensitivity. These results highlight an interesting relationship between basic decision-making research and the study of the response to traumatic events.

Private e-mail requests and the diffusion of responsibility☆

E-mail technology provides a way of requesting information or assistance from multiple sources by simultaneously addressing a letter to more than one recipient. Models of prosocial behavior taken from social psychology and economics suggest that the probability of receiving a helpful response is an inverse function of the number of simultaneous addressees. An experiment is presented which examines this prediction in the context of an e-mail request for information. The results show that there are more responses to e-mails addressed to a single recipient, that these responses are more helpful, and that they are lengthier. Response rates and measures of helpfulness were found to be independent of explicit information pertaining to the ability of other recipients to provide assistance. Implications of the results for the application of social cueing theory to e-mail communication and direct marketing are discussed.

The role of information sampling in risky choice

Life is a gamble. True to the cliche, we can rarely be certain of the consequences of our everyday actions. Presumably because of lifes lack of certainty, much of the psychological research on decision making under risk is devoted to the study of choices between monetary garnbles or prospects. With few exceptions, the outcomes and outcome probabilities of such garnbles are explicitly described to the decision maker before he or she makes a choice. How often are lifes gambles described in this way? When we decide whether to back up our computers hard drive, cross a busy street, or go out on a date, we often do not know the cornplete range of the possible outcomes, let alone their probabilities. Yet we routinely rnake such decisions, usually without difficulty. We do so by drawing upon our experience with the relevant prospects, for instance, by recalling the outcornes of previous choices. Though effective, sampling from experience brings with it the potential to generate a skewed picture of the risky prospects we face. As we will argue here, this potential is greatest when the outcome probabilities are small and relatively few outcomes have been experienced by the decision maker. Under these circumstances, decisions derived from experience are likely tobe systematically different from those rnade in full knowledge of the outcomes and outcome probabilities. In this chapter, we investigate risky decisions made on the basis of sampled experience. Three findings deserve emphasis. First, when outcome information is sampled, the likelihood of rare events is often underestimated. Second, when the sampling process is sequential, more recent outcomes tend to receive more weight than do earlier outcomes. Third, in decisions from experience, the option that appears best in light of the experienced sample tends to be the one selected. To account for these effects, we propose a learning mechanism that models how people update their estimates of a risky prospects value in light of newly sampled outcomes.

Predictive value and the usefulness of game theoretic models

Abstract Green [Int. J. Forecasting (2002)] reports that in certain settings predictions made by game theorists can be outperformed by the outcome of a short role playing exercise. Goodwin [Int. J. Forecasting (2002)] argues that this does not imply that game theoretic analysis cannot be useful. The current paper discusses two types of observations that support this assertion. First, there are many important settings in which game theoretic models have high forecasting power. Two examples: the aggregate outcome of entry job markets, and the outcome of repeated interactions are summarized here. The second observation concerns the possibility of objectively forecasting the predictive value of specific models (and methods) on particular domains. To increase our understanding of the value of role playing, we suggest that future research focus on estimating the predictive value of this method using a random selection of problems from a well defined set.

Learning to Ignore Online Help Requests

Email discussion groups provide a useful way of organizing email communities with a common interest in a certain topic. Emails submitted to the discussion group are sent automatically to each individual member, thereby eliminating the need to send multiple emails. This method may present unexpected difficulties however, when it comes to cooperation between members. An experimental study shows that email requests for help sent through discussion groups received less responses than emails sent individually to members of a group. Furthermore, subscribers to large discussion groups responded less often to help requests, whether they were sent to the group as a whole or to individuals. These results are discussed in terms of the separate roles of social cues and experience on the diffusion of responsibility effect.

Underweighting rare events in experience based decisions: Beyond sample error

Recent research has focused on the “description-experience gap”: while rare events are overweighted in description based decisions, people tend to behave as if they underweight rare events in decisions based on experience. Barron and Erev (2003) and Hertwig, Barron, Weber and Elke (2004) argue that such findings are substantive and call for a theory of decision making under risk other than Prospect Theory for decisions from experience. Fox and Hadar (2006) suggest that the discrepancy is due to sampling error: people are likely to sample rare events less often than objective probability implies, especially if their samples are small. A strand of papers has responded examining the necessity of sample error in the underweighting of rare events. The current paper extends the results of such contributions and further strengthens the evidence on underweighting. The first experiment shows that the discrepancy persists even when people sample the entire population of outcomes and make a decision under risk rather than under uncertainty. A reanalysis of Barron and Erev (2003) further reveals that the gap persists even when subjects observe the expected frequency of rare events. The second experiment shows that the gap exists in a repeated decision making paradigm that controls for sample biases and the “hot stove” effect. Moreover, while underweighting persists in actual choices, overweighting is observed in judged probabilities. The results of the two experiments strengthen the suggestion that descriptive theories of choice that assume overweighting of small probabilities are not useful in describing decisions from experience. This is true even when there is no sample error, for both decisions under risk and for repeated choices.

Combining a Theoretical Prediction with Experimental Evidence

We propose a novel experimental design to assess how to combine predictions from a theoretical model with experimental evidence to yield new, more accurate quantitative predictions. The first step involves deriving the predictions of the theoretical model by estimating unobserved parameters. The second step involves estimating the optimal weights with which to combine the theoretical predictions with experimental evidence. This latter estimation uses a random sample of experimental tasks from the relevant space. The optimal weight to give to the theoretical prediction can be summarized by a single number - the Equivalent Number of Observations (ENO). This number has an intuitive interpretation: the prediction of the theory is as accurate as the prediction from an experiment with n = ENO observations of the task to be predicted. To demonstrate and evaluate the use of the equivalent number of observations, the paper examines popular models in two of the best-studied problems in experimental economics: individual decision-making under uncertainty and repeated play of constant sum games. The two examples show that in addition to solving applied problems, the results provide some insights concerning the interpretation of descriptive models as useful approximations.

Right of way in the sky: Two problems in aircraft self-separation and the auction-based solution

There has been a growing movement to give commercial airliners more freedom in choosing their routes and responsibility for detecting and avoiding conflicts. These “free flight” concepts must contain new rules for assigning right of way in potential conflict situations. To evaluate the effect of prospective rules, the current paper derives the expected response of agents who exhibit different levels of sophistication. Traditional game theoretic analysis is used to derive the behavior of rational agents. Computer simulations are used to predict the behavior of boundedly rational reinforcement learners. The results reveal that several seemingly reasonable, straightforward right-of-way rules might lead to undesirable outcomes. These problematic results are robust to the assumed level of rationality. It is shown that these problems can be alleviated by using auctions to resolve competition for right of way. Actual or potential applications of this research include the usage of second price auctions to address right-of-way and similar conflicts