Psychology of Decision Making: recent developments
University of Hertfordshire
Convenor & Chair: Ken Gilhooly, School of Psychology, University of Hertfordshire
Discussant: Daniel Kahneman, Princeton University, USA
This symposium focuses on recent research on the psychology of decision making which builds on the pioneering contributions of Nobel prize winner in Economics (2002) and symposium discussant, Daniel Kahneman. Kahneman’s work, with the late Amos Tversky, has cast light on short-cutting heuristics and biases that account for differences between actual human decision making and that which would be expected from the idealised rational decision makers traditionally invoked in economic theory.
The papers in the present symposium contribute to our growing understanding of processes underlying decision making. Harvey and Reimers explore the ‘Anchoring and adjustment’ heuristic (initially described by Kahneman & Tversky) and its effect on under-prediction of trends by forecasters. Stewart presents a new theoretical model for risky decision making, which is compared with Kahneman and Tversky’s prospect model and others against a range of data sets. Read and Grushka-Cockayne examine the similarity heuristic, which is derived from Kahneman and Tversky’s representative heuristic and argue from data that the similarity heuristic is a type of ‘fast and frugal’ heuristic. Ayton and Pesola develop and test the notion that subjective value for a given amount of money depends on its relation to a reference level. Satisfaction with amount of money won in a realistic game was affected by comparison with how much might have been won if different choices had been made.
The anchoring and adjust heuristic: Insights from forecasting research
Nigel Harvey, Stian Reimers, University College London
Objectives: Anchoring and adjustment was one of the three heuristics proposed by Daniel Kahneman and Amos Tversky in their original 1974 paper. People start from an initial value and adjust away from it to produce their judgment. The initial value is given or suggested by the situation. Typically, adjustments are insufficient: this produces the anchoring bias. Various models of the cognitive processes responsible for implementing the heuristic have been developed to explain this under-adjustment. Tversky and Kahneman (1974) stressed that ‘heuristics are highly economical and usually effective’. While the effectiveness of availability and representativeness in many situations is relatively easy to appreciate, the adaptive value of using the anchor-and-adjust heuristic is rather harder to understand. Why is a heuristic that systematically produces a bias in one direction effective? Given feedback from the environment, how does under-adjustment survive? We use some of our recent results from experiments on judgmental forecasting from time series data to suggest a solution to this puzzle. Anchoring-and-adjustment has been used to explain why experiments have shown that forecasters underestimate of the steepness of trends (trend-damping). On average, the last point is on the trend line; it acts as a judgment anchor. People adjust from this point to take the trend into account. Because they under-adjust, trend-damping occurs. Interestingly, trend-damping appears with simulated series (for which the structure is known) but not for real series (for which the structure is uncertain). Furthermore, statisticians developing formal methods for forecasting have found that including a damping term in their formulae improves accuracy. Outside the laboratory, linear trends tend to be part of long-term cycles. Thus people may damp trends because of their experience within their ecology. This suggests that experience with different types of series can act to modify trend-damping.
Design: In two of our experiments, two series had the same trends in two conditions but trends varied across these conditions in the remaining series. We expected degree of trend-damping in the two series with trends constant across conditions to change as trends in the remaining series varied. In a third experiment, we checked that damping was not purely dependent on contextual effects produced within the laboratory by carrying out a single-shot experiment on a large number of people.
Method: Series were presented graphically in web-based experiments. People made a sequence of forecasts from each series.
Results: We obtained the predicted contextual effects on degree of trend-damping in our first two experiments. A large trend-damping effect was obtained in our third experiment.
Conclusions: Results are consistent with the view that trend-damping is an adaptation to characteristics of the ecology. Other cases of under-adjustment that are well-documented within the judgmental forecasting literature can be explained in this way. It is likely that the anchoring-and-adjustment heuristic is also broadly adaptive outside the forecasting domain.
Risky Decisions by Sampling
Neil Stewart, University of Warwick
I extend the decision by sampling (DbS) model of risky decision making to predict choices between risky prospects. DbS assumes that three simple cognitive tools are the basis for decision making: binary, ordinal comparison; sampling; and frequency accumulation. The model is quite simple,but works surprisingly well. On each time step, a target attribute value (an amount or probability) is randomly selected from one prospect and a binary, ordinal comparison is made to a comparison attribute value randomly selected from another prospect or from long-term memory. For each prospect, a frequency accumulator tallies the number of favourable binary, ordinal comparisons. The prospect with the higher tally is chosen when the difference between accumulator tallies reaches a threshold.
This DbS model is compared with existing models derived from the dominant expected utility framework including cumulative prospect theory (Kahneman & Tversky, 1979), decision field theory (Busemeyer & Townsend, 1993), and the transfer-of-attention-exchange model (Birnbaum & Chavez, 1997) as well as the priority heuristic model (Brandstaetter, Gigerenzer & Hertwig, 2006). Models are tested in competition on a series of benchmark data sets.
The Similarity Heuristic
Daniel Read, Durham Business School, Yael Grushka-Cockayne, London Business School
The heuristics and biases program initiated by Kahneman and Tversky has been subject to attack based on a purported failure to operationalise the heuristics, and on the use of systematic rather than representative experimental designs. In this paper one part of the original representativeness heuristic, which we call the similarity heuristic, is operationalised and tested with a representative design. The similarity heuristic is what we do when we decide to act based on our subjective judgments of similarity. For instance, if we decide whether or not to eat mushroom based on whether it looks poisonous or non-poisonous, we are using the similarity heuristic. We describe a theoretical and empirical investigation of this heuristic. We start with a mathematical model which permits quantitative analyses of its validity (when it will be successful, and when it will not), and precise predictions of choice behaviour. This model is tested with an experiment showing that, at least in the environment we investigate, similarity judgments can be a reliable and valid basis for choice, and, crucially, the similarity heuristic is also used to make choices. Similarity judgments made by one group almost perfectly predicted choices made by another group. We also show an important characteristic of how people use prior probability (or base rate) information – while people do use the information, they do not use it strategically. Knowledge of priors appears to act as a bias that increases the likelihood of choosing he high prior item, but does not necessarily increase accuracy. We argue that the similarity heuristic is one of a class of ‘fast and frugal’ heuristics (Gigerenzer & Goldstein, 1996) that yield good results at low cognitive cost.
Risky Decisions and Perceived Happiness of ‘Deal or No Deal’ TV game show contestants
Peter Ayton, Meri Pesola, Department of Psychology, City University
Objectives: The notion that people do not classify or judge perceived objects independently of their context has considerable economic significance when it comes to the human reaction to money. The idea that people’s utility for money is dependent on its relation to a reference level follows from psychological theory (cf. Tversky and Kahneman, 1992) and receives support from diverse empirical sources (e.g. Clark & Oswald, 1996). Here we explored how the rated happiness of TV game show contestants varies as a function of their monetary winnings and the money that they believe that they could have won if they had dealt differently. The TV game show enables us to study the effects on emotions of outcomes from simple well-defined decisions involving real large stakes not used in experiments. ‘Deal or No Deal’ is a game of chance – participants select a series of boxes to reveal which of a range of amounts of money varying from 1p to £250,000 is not in their box. At several stages the ‘Banker’ offers contestants a riskless amount of money to quit the game: contestants can ‘deal’ and quit or play on. When and if they deal they must indicate which boxes they would have chosen – putatively to see what would have happened if they had not dealt.
Design: We presented respondents with descriptions of the sequence of events of individual contestants, including the deals they were offered both before and after the contestant dealt and the decisions they made at each stage and the amount of money each contestant won. Respondents were asked to rate how happy they thought the contestant was, how happy they would be and how much the contestant might feel regret – all at the end of each game.
Method: 200 game shows were rated by 80 participants who completed the ratings task via a web page. Each participant rated the happiness of ten contestants. Between participant reliability of ratings was satisfactory.
Results: Analyses of the happiness ratings revealed that, unsurprisingly, participants perceptions of the happiness of contestants were influenced by the amounts of money that contestants won. However the perceived happiness of contestants was strongly moderated by the magnitude of the discrepancy between their winnings and the highest offer that they declined; contestants who declined offers higher than their eventual winnings were less happy than those who declined offers lower than their eventual winnings. Contestants who dealt and then discovered that they would have been offered more had they not dealt and played on were also less happy than those who dealt and discovered their future offers would have been less than their winnings.
Conclusions: This study confirms the idea that satisfaction is not a simple function of the absolute values of outcomes. Happiness with amount of money won is affected by comparison to how much money might have been won if different choices had been made.