Saturday, October 12, 2013

Lecture Summary: Judgement, Heuristics and Biases

I am currently giving a set of lectures as part of a module "Behavioural Economic: Concepts and Theories" in Stirling. I am posting brief informal summaries of some of these lectures on the blog to generate discussion. Thanks to Mark Egan for a lot of help in putting these together online. 

This lecture covered Judgement, Heuristics and Biases and gave an overview of how this topic relates to textbook models of choice under uncertainty and value. The lecture outlines what is meant by heuristics; discusses theories of why they may be adaptive; outlines the idea of bias and how this relates to heuristics; examines the main Kahneman and Tversky literature and looks at examples of experiments on availability, representativeness and anchoring; examines overconfidence and planning fallacy; examines the role of emotions in guiding probability judgments and risk perception; discusses limitations of the literature on heuristics and biases; discusses the implications of use of heuristics for economic theory and policy.

What are heuristics?
This article by Kahneman and Tversky (1974) is still a classic description of the main heuristics that people use to judge probability and frequency. Heuristics can be thought of as mental 'rules of thumb' that people employ for all kinds of judgements. For example, if you want to share a cake among 5 people, rather than optimize the size of each slice depending on each person's unique preferences, level of hunger, etc you might employ a 1/n heuristic and give everyone an equal 1/5th slice. Or if you see dark clouds forming on your way to work, you might decide to bring a raincoat. To paraphrase Kahneman & Tversky, "People rely on heuristic principles to reduce the complex tasks of assessing probabilities to simpler judgmental operations." I'll briefly discuss some experiments and examples about the three heuristics from this paper.

a. Anchoring Heuristic:
Making estimates by starting from an initial value that is then adjusted to get the final answer. The adjustment is usually insufficient.

Fig 1. Anchoring
Participants were asked to estimate various quantities in percentages - for example what percentage of countries in the UN are from Africa?. Before answering, participants watched as a (rigged) wheel of fortune was spun to produce either the number 10 or 65. They were then asked whether the answer to the UN question was higher or lower than the number on the wheel of fortune. They then gave their estimate by moving up or down from that number. 

The arbitrary numbers shown on the wheel of fortune had a marked effect on estimates. The median estimates of the percentage of African countries in the United Nations were for 25% and 45% for groups that received 10 and 65, respectively, as starting points. Payoffs for accuracy did not reduce the anchoring effect.

Example: This paper by Ariely et al. (2004) is important as it documents how people are influenced by the anchoring heuristic when making judgements of the value of goods. I will spend a little time picking on the people who are reviewing this article for their journal assignment in terms of asking them questions on this.

b. Availability Heuristic:
Judging the probability of an event or the frequency of a class based on the ease with which occurrences or instances can be brought to mind.

Participants heard a list of well-known personalities of both sexes and were subsequently asked to judge whether the list contained the names of more men or more women. Different lists were presented to different groups of subjects. In some lists the men were relatively more famous than the women, and in others the women were relatively more famous than the men.  In each of the lists, the participants wrongly concluded that the class (sex) that had the more famous personalities was the more numerous.

Example: When people are asked whether it is more likely that a word randomly sampled from an English text begins with ‘r’ or has ‘r’ as its third letter, they begin by recalling words that begin with ‘r’ (road) and words that have ‘r’ in the third position (e.g. car).  Because it is easier to mentally search for words by their first letter than by their third letter, most people judge words that begin with a given consonant to be more numerous than words in which the same consonant appears in the third position. This is true even for consonants, such as ‘r’ or ‘k’, which are more frequent in the third position than in the first

c. Representativeness Heuristic:
Probabilities are evaluated by the degree to which A is representative of  B -  by the degree to which A resembles B. For example, when A is highly representative of B, the probability that A originates from B is judged to be high. On the other hand, if A is not similar to B, the probability that A originates from B is judged to be low.

Participants were shown brief personality descriptions of several individuals, supposedly sampled at random from a group of 100 engineers and lawyers. An example description was "Steve is very shy and withdrawn, invariably helpful, but with little interest in people, or in the world of reality. A meek and tidy soul, he has a need for order and structure, and a passion for detail."

For each description like the one above, participants were asked to judge the probability that it belonged to an engineer rather than to a lawyer. In one condition, subjects were told that the group from which the description had been drawn consisted of 70 engineers and 30 lawyers. In another condition, subjects were told that the group consisted of 30 engineers and 70 lawyers.  The probability that the description belonged to an engineer rather than a lawyer should therefore be higher in the first condition. Regardless of the condition, the subjects produced the same probability judgements. The subjects evaluated the probability that the description referred to an engineer rather than a lawyer based on the description’s representativeness to the two stereotypes and did not take into account the prior probabilities of each category. 

Example: In considering tosses of a coin for heads or tails, people judge the sequence H-T-H-T-T-H to be more likely than the sequence H-H-H-T-T-T, which does not appear random, and more likely than the sequence H-H-H-H-T-H, which seems unfair. People expect that the essential characteristics of the process will be represented not only globally in the entire sequence, but also locally in each of its parts. A locally representative sequence, however, deviates systematically from chance expectation: it contains too many alternations and too few runs. After observing a long run of red on the roulette wheel, most people falsely believe that  black is now due, presumably because the occurrence of black will result in a more representative sequence than the occurrence of an additional red.

Scope Neglect 
Fig 2. Willingness to pay to save X number of birds
Def: An inability to scale valuation of a problem as its size increases.

Desvousges et al. (1993) asked participants in an experiment this question: 

"(2,000 / 20,000 / 200,000) migrating birds die each year by drowning in oil ponds. These deaths could be prevented by covering the oil ponds with nets. 

How much would you be willing to pay to provide the nets?

The results (Fig. 2) found a non-linear increase in willingness to pay as the size of the problem increased. People were willing to spend $80 to save 2,000 birds, but only $8 more to save 200,000 birds. The unfortunate problem is that although there are 100x more birds at risk, we do not feel 100 times more alarmed or upset. See #48 here for more work on this area.

Risk Perception and Public Policy 
It is worth thinking about this in the context of terror alerts about potential Mumbai-style attacks in Europe. I am going to give the people who issue these types of alerts the benefit of the doubt and assume that there is a good reason for doing them e.g. to allow people to make their own choice, to raise general alertness about threats to aid prevention and so on. However, it does raise the strong possibility that people will start worrying about things with very low probabilities of occurring that might affect them and over which they do not have much control. Furthermore, there is a huge imbalance in the airtime devoted to making people afraid of things compared to the airtime devoted to getting people to set things in their proper context e.g. there are terrorist threats in Europe but you are statistically far more likely to be killed in a car accident.

Given the probabilities involved for any individual in dying in a terrorist attack while travelling to Continental Europe, my response to the class was that if they intended cancelling an upcoming trip to France or Germany then I suggested that we lock the doors and I will call up for supplies and we will barricade ourselves in the lecture theatre because the relative safety of staying there for the rest of the term compared to braving the roads home everyday is far greater than the relative safety of staying at home for a few days compared to taking your trip. It is easier to imagine dying in a terrorist attack than dying in a car crash and it occupies more of our mind when we are recalling risk. It is not at all inconsistent to feel moral outrage about terrorist attacks and seek that they be prevented and/or the perpetrators pursued but it is inconsistent with a rational worldview to overestimate your exposure to risk and to allow this to influence your decisions. Part of living in the modern world is overcoming fears of salient but low probability and non-catastrophic risk and getting on with life and making decisions based on more core values. Given the huge asymmetric costs faced by public officials from not revealing information about risk, it may ultimately have to be up to us as individuals to learn how to filter out things that lead to wrong decisions and a life based on fear.

The point of the Beshears et al, paper mentioned in the previous lecture is that if people are subject to all the various psychological constraints that we will discuss in this course, then their behaviour in markets is no longer the best method of judging their "real" preferences. We then need to ask how do we know what people want and how can this influence society and public policy. I know some of you are baffled as to why economists could believe the model of choice set out in the first part of the lecture. It is worth remembering that they do not necessarily believe that (a) people always behave like that and (b) they are not subject to all manner of biases. What many theorists believe is that these assumptions do a good enough job at characterizing real-world behaviour in important contexts and that including all of these other features of decision making leads to a model of society and the economy that is too complex to provide meaningful input into institutional design. 


Fig 3. The Rational Molasses Model
Herbert Simon, in his 1959 paper "Theories of Decision-Making in Economics and Behavioural Science", discusses why the rational actor model is valuable and useful despite the complexities of human psychology, by way of an analogy of pouring sticky molasses into a irregularly shaped bowl. He writes "What would we need in order to make a theory of the form the molasses would take in the bowl? If the bowl were held motionless, and if we wanted only to predict behavior in equilibrium, we would have to know little, indeed, about molasses. The single essential assumption would be that the molasses, under the force of gravity, would minimize the height of its center of gravity. With this assumption...and a complete knowledge of the environment - in this case the shape of the bowl - the equilibrium is completely determined. Just so, the equilibrium behaviour of a perfectly adapting organism depends only on its goal and its environment: it is otherwise independent of the internal properties of the organism." Simon goes on to note that if the bowl was being jiggled rapidly, or if we wanted to know about the behaviour of the molasses before equilibrium, then we would require much more information to make a prediction - the precise viscosity of the molasses, how quickly it "adapted" to the shape of the bowl, how quickly it moved towards its "goal" of lowering its center of gravity, etc. With this analogy Simon wants to make the point that making predictions of short-run economic behaviour quickly becomes exceedingly complex, potentially demanding an extreme level of detail, but you might well do a reasonable job of predicting long-term behaviour by focusing only on the essential characteristics of the decision-making process.

To further emphasize the point, when the behavioural economist Richard Thaler was asked "When will there be a single unified "behavioral" theory of economic activity?" he answered:

Fig 4. Richard Thaler

"Never. If you want a single, unified theory of economic behavior we already have the best one available, the selfish, rational agent model. For simplicity and elegance this cannot be beat. Expected utility theory is a great example of such a theory. Von Neumann was no dummy! And if you want to teach someone how to make good decisions under uncertainty, you should teach them to maximize expected utilityThe problem comes if, instead of trying to advise them how to make decisions, you are trying to predict what they will actually do. Expected utility theory is not as good for this task. That is the purpose for which Kahneman and Tversky's descriptive alternative, prospect theory, was invented."

From a broader philosophical perspective, many have worried that the intrusion of psychology into economic design leads to the debasement of human freedom by the increasing justification of greater state control of behaviour. There will be a wide variance of views in the class on these issues from people who believe that the state should play far more of a role in achieving positive outcomes to those who believe that the state should be expunged from these functions. Whatever your view (and keep an open-mind) it makes things far more complex than simply asking whether we believe in the validity of, for example, the Kahneman and Tversky heuristics experiments. We are also asking whether these influences are sufficiently important that they survive the move from lab to real-world, important, repeated interactions and whether they have important consequences for institution design. In some sense, the discipline of behavioural economics is an extreme attempt to establish the ecological validity of cognitive and social psychology.

Questions and Follow-Ups 

A few of you were asking about the relationship between individual rationality and wider institutional and ideological analysis. I will talk a lot about this later. I was giving a talk with a group of people who come at things from a very institutionalist perspective and asked them for the most interesting papers that would link institutional analysis to behavioural economics. This paper on "mental models" by Douglas North and Arthur Denzau is fascinating reading on that score. It is not required reading for the course but they will be of interest to people interested in the link between broader societal and institutional analysis and the topic of individual rationality.

Another question that was asked was the extent to which full information optimising behaviour is actually rational. In other words, would it not be more rational to acknowledge that we have limitations and then use rules of thumb to help us make decisions more swiftly. We will be talking about bounded rationality and heuristics in later lectures. This article by Geoffrey Hodgson argues that habits and rules are pervasive features of human action in a wide variety of domains. Again, not required reading but fascinating if you are interested in this area. Some students asked about the public policy implications of heuristics and biases. There are dozens of ways you could attack this aspect. Let me give you some ideas that I covered at length in the lectures and hopefully you will see how they link. If you have specific other ideas you are looking at please feel free to email me. One way to examine whether there are policy implications of the evidence on biases and heuristics  is to examine why the presence of biases is important. The main article that looks at the implications of things like judgement biases is the Beshears et al article. You will have to read this yourself and flesh out your answer. But their basic point is that if people make mistakes in decisions (e.g. by being confused, misjudging risk etc.,) then their decisions may no longer be a good guide to their actual welfare. This would imply that the market system itself may not necessarily yield the best outcomes for consumers. It also potentially implies that giving consumers more information, framing that information more clearly etc., might improve their welfare. Thus the main public policy implication would be that it might be possible to improve people's welfare through some of these mechanisms, something that would be explicitly ruled out by a model where people fully and rationally interpret risk.

Similar to this is the question of whether the results that have been found are "important" enough to be taken seriously for public policy. These effects are very interesting but that is different from being important as it could be the case that something found in small-scale experiments with trivial examples might not extend to real-world and important decisions. With that in mind, the Ariely paper on the biases and heuristics reading list shows that, for example, anchoring occurs even when the goods are consumer goods rather than just random questions like in the original Kahneman and Tversky article. Also, the Gigerenzer paper that I spoke about indicates that many people lost their lives after 9-11 because they switched to driving, wrongly believing due to the availability heuristic that driving was safer than flying. If judgement biases are leading to large death tolls like this, then there are obviously many good policy reasons to try to take them into account when communicating risk.

Readings 
1. Kahenman & Tversky (1974), Judgment under Uncertainty: Heuristics and Biases, Science. 
2. Ariely, Loewenstein & Prelec (2004), “Coherent Arbitrariness”: Stable Demand Curves Without Stable Preferences, QJE
3. North & Denzau (1994), Shared Mental Models: Ideologies and Institutions, Kyklos
4. Gigerenzer (2004), Dread Risk, September 11, and Fatal Traffic Accidents, Psychological Science
5. Hodgson (1997), The ubiquity of habits and rules, Cambridge Journal of Economics
6. Beshears et al. (2008), How are Preferences Revealed, NBER Working Paper

No comments: