NOTES - Gut Feelings: The Intelligence of the Unconscious - Gerd Gigerenzer

Gut Feelings: The Intelligence of the Unconscious - Gerd Gigerenzer (2007)



Chapter 1: Gut Feelings

1. Franklin, 1779. The scientist and statesman Benjamin Franklin was one of the great figures of the Enlightenment, and his moral algebra is an early version of modern utilitarianism and rational choice theory. In his ethics, the rake and the drunkard are just like other people but became that way simply because they failed to calculate correctly.

2. Wilson et al., 1993. Similarly, Halberstadt and Levine, 1999, and Wilson and Schooler, 1991, experimentally showed that introspection can reduce the quality of decisions, and Zajonc, 1980, and Wilson, 2002, provided further stories on the conflict between balance sheets and gut feelings. Wilson reports of a social psychologist who tried to decide whether to accept a job offer from another university by using a balance sheet, listing pros and cons. Halfway through, she said, “Oh hell, it’s not coming out right! Have to find a way to get some pluses on the other side” (167).

3. Schwartz et al., 2002. The term satisficing was introduced by the Nobel laureate Herbert A. Simon and originated in Northumbria, a region in England on the Scottish border, where it means “to satisfy.”

4. Goldstein and Gigerenzer, 2002. The term heuristic is of Greek origin and means “serving to find out or discover.” The Stanford mathematician G. Polya, 1954, distinguished between heuristic and analytic thinking. For instance, heuristic thinking is indispensable for finding a mathematical proof, whereas analytic thinking is necessary for checking the steps of a proof. Polya introduced Herbert Simon to heuristics, and it is on the latter’s work that I draw. Independently, Kahneman, et al., 1982, promoted the idea that people rely on heuristics when making judgments but focused on errors in reasoning. In this book, I use heuristic and rule of thumb as synonyms. A heuristic, or rule of thumb, is fast and frugal; that is, it needs only minimal information to solve a problem.

5. Dawkins, 1989, 96.

6. Babler and Dannemiller, 1993; Saxberg, 1987; Todd, 1981.

7. McBeath et al., 1995; Shaffer et al., 2004.

8. McBeath et al., 2002; Shaffer and McBeath, 2005.

9. Sailors learn that if another boat approaches, and the bearing remains constant, a collision will occur. Soldiers are taught that if mortar or shells are fired at them, they should wait until the object is high up in the air and then point at it. If it does not move relative to their finger, they’d better run. If the object descends below their finger, it is going to land in front of them, and if it continues to ascend, it will land behind them.

10. Collett and Land, 1975; Lanchester and Mark, 1975.

11. Shaffer et al., 2004.

12. Dowd, 2003.

13. Horan, in press.

14. Lerner, 2006.

15. Commercial egg producers want to identify female chickens quickly and avoid feeding the unwanted sex: the unproductive males. Before the art of chicken sexing emerged in Japan, poultry owners had to wait until chicks were five to six weeks old; today, expert chicken sexers can reliably determine the sex of day-old chicks on the basis of very subtle cues, at a speed of some thousand chickens per hour. R. D. Martin, author of The Specialist Chick Sexer (1994), cites an expert on his publisher’s Web site: “There was nothing there but I knew it was a cockerel” and comments, “This was intuition at work.” Like other tacit skills, sexing can become an obsession. “If I went for more than four days without chick sexing work I started to have ‘withdrawal symptoms.’”

16. Yet this belief is alive and kicking. Even when it comes to emotional intelligence, it is still assumed that it can be measured by asking people questions concerning declarative knowledge, for instance, to rate themselves on the statement “I know why my emotions change” (see Matthews, et al., 2004). The underlying belief is that people are able and willing to tell how their intelligence functions. In contrast, the influential work by Nisbett and Wilson, 1977, reviewed experimental evidence that people often do not have introspective access to the reasons underlying their judgments and feelings. Research on implicit learning refers to learning that proceeds both unintentionally and unconsciously (Lieberman, 2000; Shanks, 2005).

17. For similar definitions, see Bruner, 1960, Haidt, 2001, and Simon, 1992.

18. See Jones, 1953, 327, on Freud, and Kahneman et al., 1982, on cognitive illusions. For my critique of these views, see Gigerenzer, 1996, 2000, 2001; for a reply to my critique, see Kahneman and Tversky, 1996, and Vranas, 2001.

19. For instance, Gladwell’s (2005) engaging book Blink features research, including my own, on how people make successful snap judgments: “and—blink!—he just knows. But there’s the catch: much to Braden’s frustration, he simply cannot figure out how he knows” (49). In this book, I try to explain how these intuitions work.

20. Wilson, et al., 1993, 332, explained why women who gave reasons were less satisfied with the posters they chose in this way: “Introspection…can change an optimal weighing scheme into a suboptimal one. When people analyze reasons, they might focus on those attributes of the attitude object that seem like plausible causes of the evaluations but were not weighted heavily before.” The idea that the process underlying intuitive judgments resembles Franklin’s balance sheet and rational choice theory is also found in Dijksterhuis and Nordgren, 2006, and Levine et al., 1996. These distinguished researchers demonstrate in their fascinating experiments that less thought can be more. To explain this phenomenon, however, they do not push the vision that less could actually be more (see next chapters) but instead assume that instantaneous judgments, if they are good, must be based on unconscious calculations of all pros and cons.

21. This includes Ambady and Rosenthal, 1993, Cosmides and Tooby, 1992, Gazzaniga, 1998, Hogarth, 2001, Kahneman et al., 1982, Myers, 2002, Payne et al., 1993, Pinker, 1997, and Wegner, 2002. For an introduction to the research at the Max Planck Institute see Gigerenzer et al., 1999, Gigerenzer and Selten, 2001, Gigerenzer, 2004a, and Todd and Gigerenzer, 2003.

Chapter 2: Less Is (Sometimes) More

1. Einstein is quoted in Malkiel, 1985, 210. For Einstein, the simplicity of an explanation was both a sign for its truth and a goal of science: “Our experience up to date justifies us in feeling sure that in Nature is actualized the ideal of mathematical simplicity” (Einstein, 1933, 12).

2. Bursztajn et al., 1990.

3. Luria, 1968, 64.

4. Anderson and Schooler, 2000; Schacter, 2001; Schooler and Hertwig, 2005.

5. James, 1890/1981, 680. The Word buffer analogy to memory retrieval in Figure 2-1 was proposed by Lael Schooler and is based on Schooler and Anderson, 1997.

6. That is, it was unable to pick up concepts such as noun-verb agreement in embedded clauses (Elman, 1993; see also Newport, 1990).

7. Cited in Clark, 1971, 10.

8. Huberman and Jiang, 2006.

9. DeMiguel et al., 2006. The optimal asset allocation policies included sample-based mean-variance portfolios, minimum-variance portfolios, and strategies for dynamic asset allocation. These policies based their estimates on ten years of past financial data and had to forecast the performance in the following month. For similar results, see Bloomfield et al., 1977. The 1/N rule is a version of equal-weight or tallying rules, which have been shown to match or outperform complex weighting policies in speed and accuracy (Czerlinski et al., 1999; Dawes, 1979). See Zweig, 1998, on Markowitz.

10. Ortmann et al., in press; Barber and Odean, 2001.

11. We used the 500 Standard & Poor stocks and 298 German stocks and asked four groups—Chicago pedestrians, University of Chicago students of business, Munich pedestrians, University of Munich students of business—which of the stocks they recognized by name (Borges et al., 1999). We then constructed eight high-recognition portfolios (U.S. and German stocks, for each of the four groups) and evaluated their performance after six months against four benchmarks: market indices, mutual funds, random (“dartboard”) portfolios, and low-recognition portfolios. The high-recognition portfolios outperformed the respective market indices (Dow 30 and Dax 30) and the mutual funds in six out of eight cases and matched or outperformed random portfolios and low-recognition portfolios in all cases. This study prompted a great deal of press coverage and two opposite reactions. There were those, notably financial advisers, who said, “It can’t be true,” and those who said, “No surprise. We knew it all along.” One objection was that we had hit a bull market, yet in two subsequent studies, we hit a bear market and could still replicate the beneficial effect of name recognition (Ortmann et al., in press). However, two other studies did not report name recognition to be of advantage, each of these relying exclusively on the recognition of college students rather then the general public. Boyd, 2001, used students whose recognition of stocks was idiosyncratic and resulted in disproportionate losses or disproportionate gains. Frings et al., 2003, excluded all students who recognized more than 50 percent of the Nemax50 and violated the diversification principle (at least ten stocks in one portfolio) used in our studies. All in all, the studies suggest that portfolios built on collective brand-name recognition perform about as well as and sometimes better than financial experts, the market, and mutual funds do.

12. Sherden, 1998, 107. For instance, between 1968 and 1983, the market outperformed pension fund managers by about .5 percent per year. Adding on management fees, this results in-1 percent per year. In 1995, the Standard & Poor’s Index rose by 37 percent, while mutual funds increased by only 30 percent, and the majority (89 percent) could not beat the market. See also Taleb, 2004.

13. Goode, 2001.

14. This magical number was proposed by the psychologist George A. Miller, 1956. Consistent with this number, Malhotra, 1982, concludes that in consumer decisions, ten or more alternatives cause poorer choices.

15. Iyengar and Lepper, 2000.


17. Lenton et al., 2006.

18. Beilock et al., 2004; Beilock et al., 2002.

19. Johnson and Raab, 2003. The I-shaped bars in Figure 2-3 are standard errors of means.

20. Klein, 1998.

21. Wulf and Prinz, 2001.

22. See Carnap’s (1947) “principle of total evidence” and Good’s (1967) “total evidence theorem,” both of which state that information should never be ignored, and Sober, 1975, for a discussion. Hogarth, in press, reviews four areas in which simple strategies consistently outperform those that use more information: simple actuarial methods predict better than sophisticated clinical judgment; simple methods in time-series forecasting are superior to “theoretically correct” methods; equal weighting (tallying) is often more accurate than using the “optimal” weights; and decisions can often be improved by discarding relevant information. Hogarth concludes that in each of these domains, demonstrations that simple strategies are better than complex ones at predicting complex phenomena have been largely ignored, since the idea is difficult to accept by most researchers. On less is more, see Hertwig and Todd, 2003.

Chapter 3: How Intuition Works

1. Cited in Egidi and Marengo, 2004, 335. Whitehead was an English mathematician and philosopher who coauthored Principia Mathematica with Bertrand Russell.

2. Darwin, 1859/1987, 168.

3. The phrase is credited to Jerome Bruner, but the idea is older. For instance, the psychologist Egon Brunswik spoke of vicarious functioning, and Hermann von Helmholtz spoke of unconscious inferences (see Gigerenzer and Murray, 1987).

4. von Helmholtz, 1856-66/1962. In a fascinating series of experiments, Kleffner and Ramachandran, 1992, analyzed in detail how shape is inferred from shading. Bargh, 1989, provided an excellent discussion of automatic processes in general.

5. Baron-Cohen, 1995. Tomasello, 1988, showed that even eighteen-month-old babies use gaze as a cue for reference.

6. Baron-Cohen, 1995, 93.

7. Here and in the following I rely on Sacks, 1995, 259, 270.

8. Rosander and Hofsten, 2002.

9. Barkow et al., 1992; Daly and Wilson, 1988; Pinker, 1997; Tooby and Cosmides, 1992.

10. Cacioppo et al., 2000.

11. For various cognitive theories that investigate the relation between mind and environment, see Anderson and Schooler, 2000, Cosmides and Tooby, 1992, Fiedler and Juslin, 2006, and Gigerenzer, 2000.

12. Axelrod, 1984, used the prisoner’s dilemma, a strategic game that has occupied the social sciences for the last half century. Tit for tat was submitted by the psychologist and game theorist Anatol Rapaport. It tends to fail in social environments where people have the option to “quit” in addition to either being kind or being nasty (Delahaye and Mathieu, 1998). Formally, tit for tat involves players making simultaneous moves, but I also use it in a more general sense in which people can also behave successively.

Chapter 4: Evolved Brains

1. Hayek, 1988, 68. The economist and Nobel laureate Friedrich Hayek anticipated several of the ideas I propose, including that behavior is based on rules that typically cannot be verbalized by the acting person, that behavior is contingent on its environment, and that minds do not create institutions but minds and institutions evolve together.

2. Frey and Eichenberger, 1996. The Bush quote is from Todd and Miller, 1999, 287.

3. Richerson and Boyd, 2005, 100. These authors provide a highly recommended introduction to the role of imitation in cultural learning.

4. Tomasello, 1996.

5. Cosmides and Tooby, 1992.

6. Hammerstein, 2003.

7. Freire et al., 2004; Baron-Cohen et al., 1997.

8. Blythe et al., 1999.

9. Turing, 1950, 439. The philosopher Hilary Putnam, 1960, for instance, used Turing’s work as a starting point to argue for his distinction between the mind and the brain. Following Putnam, this distinction served to combat attempts at reducing the mind to the physical brain. For many psychologists, it seemed a good basis to establish the autonomy of psychology in relation to neurophysiology and brain sciences.

10. Holland et al., 1986, 2.

11. Silk et al., 2005. The results were independently replicated by Jensen et al., 2006.

12. Thompson et al., 1997.

13. Cameron, 1999.

14. Takezawa et al., 2006.

15. Henrich et al., 2005.

16. Barnes, 1984, vol. 1, 948-49. The argument I sketch is made in more detail by Daston, 1992.

17. Cited in Schiebinger, 1989, 270-72. For Darwin’s view, see Darwin, 1874, vol. 2, 316, 326-27.

18. Hall, 1904, 561.

19. The study was conducted by psychologist Richard Wiseman, University of Hertfordshire, during the Edinburgh International Science Festival in 2005. See, for example, the BBC report (

20. Meyers-Levy, 1989. The ads I mention in the following are featured in her article.

Chapter 5: Adapted Minds

1. Simon, 1990, 7. The British economist Alfred Marshall, 1890/1920, had used this analogy earlier to ridicule the debate between supply-side and demand-side theories, comparing it to an argument about whether the top blade or the bottom blade of a pair of scissors cuts cloth.

2. Simon, 1969/1981, 65. Herbert Simon was a model of a truly interdisciplinary thinker who defies any single classification as psychologist, economist, political scientist, or one of the fathers of artificial intelligence and cognitive science. On the relation between Simon and my work, see Gigerenzer, 2004b.

3. This value results from .8 × .8 + .2 × .2 = .68. That is, the rat turns left with a probability of .8, and in this case it gets food with a probability of .8, and it turns right with a probability of .2, and it gets food with a probability of .2. On probability matching, see Brunswik, 1939, and Gallistel, 1990.

4. Gigerenzer, 2006.

5. Törngren and Montgomery, 2004.

6. Sherden, 1998, 7.

7. Bröder, 2003; Bröder and Schiffer, 2003; Rieskamp and Hoffrage, 1999.

8. Gigerenzer and Goldstein, 1996; Gigerenzer et al., 1999.

9. Czerlinski et al., 1999. This procedure is known as cross-validation. We repeated it a thousand times in order to average out particular ways of dividing the data. The technical term for the hindsight task is data fitting.

10. In the case of five cities, there are five possible first cities, for each of them four possible second cities, and so on, which results in 5 × 4 × 3 × 2 × 1 possible tours. For n cities, this results in n! tours. Some of these tours have the same length; for instance the tour “a, b, c, d, e, and back to a” has the same length as the tour “b, c, d, e, a, and back to b.” For five cities, there are five starting points that lead to a tour with the same length, and furthermore there are two directions in which the tour can go. Thus, the number of possible tours needs to be divided by 5 × 2 to result in 4 × 3 = 12 tours with different lengths. In general, this number is n!/2n = (n-1)!/2. The campaign tour problem is a version of the traveling salesman problem (Michalewicz and Fogel, 2000, 14).

11. Rapoport, 2003.

12. Michalewicz and Fogel, 2000.

Chapter 6: Why Good Intuitions Shouldn’t Be Logical

1. Cartwright, 1999, 1.

2. Tversky and Kahneman, 1982, 98. Note that here and in the following the term logic is used to refer to the laws of first-order logic.

3. Gould, 1992, 469. For more on the alleged consequences, see Johnson et al., 1993, Kanwisher, 1989, Stich, 1985.

4. The linguist Paul Grice, 1989, has studied these conversational rules of thumb.

5. Hertwig and Gigerenzer, 1999; see also Fiedler, 1988, Mellers, Hertwig, and Kahneman, 2001. Tversky and Kahneman, 1983, found an effect of a relative frequency formulation in a different problem but stuck with their logical norm (Gigerenzer, 2000). Another reason for this “fallacy” is that people may read “Linda is a bank teller” as implying that “Linda is a bank teller and not active in the feminist movement.” This may happen occasionally, but it cannot explain that the “fallacy” largely disappears when the word probable is replaced by how many.

6. Edwards et al., 2001.

7. Kahneman and Tversky, 1984/2000, 5, 10.

8. Sher and McKenzie, 2006, Cognition 101: 467-94.

9. Feynman, 1967, 53.

10. Selten, 1978, 132-33.

11. Wundt, 1912/1973. On artificial intelligence see Copeland, 2004.

12. Gruber and Vonèche, 1977, xxxiv-xxxix.


Chapter 7: Ever Heard Of…?

1. Some have argued that these differences do not reflect different processes but that recognition is just a simpler form of recall (Anderson et al., 1998).

2. Dawkins, 1989, 102.

3. Standing, 1973.

4. Warrington and McCarthy, 1988; Schacter and Tulving, 1994. Laboratory research has demonstrated that memory for mere recognition captures information even in divided-attention learning tasks that are too distracting to allow more substantial memories to be formed (Jacoby et al., 1989).

5. Pachur and Hertwig, 2006. Recognition validities for Wimbledon 2003 Gentlemen’s Singles are reported in Serwe and Frings, 2006, and for city populations in Goldstein and Gigerenzer, 2002, and Pohl, 2006. Note that the recognition heuristic is about inferences from memory, not about inferences from givens, where one could be given access to information about the unrecognized alternative.

6. Ayton and Önkal, 2005. Similarly, Andersson et al., 2005, report that laypeople predicted the outcomes of the 2002 Soccer World Cup as well as experts did.

7. Serwe and Frings, 2006. The predictions were made for a sample of 96 matches out of a total of 127 matches. The correlations between the three official ratings ranged from .58 to .74, and the correlation between the two recognition rankings was .64. The betting quotes in Wimbledon, which cannot be compared to the rankings because they are updated after each game, made 79 percent correct predictions. Results were replicated for Wimbledon 2005 by Scheibehenne and Bröder, 2006.

8. Hoffrage, 1995; see also Gigerenzer, 1993.

9. This 60 percent figure is called the knowledge validity and defined as the proportion correct when both alternatives are recognized. In contrast, the recognition validity is defined as the proportion correct when only one alternative is recognized, and the recognition heuristic is used (Goldstein and Gigerenzer, 2002).

10. The curves in Figure 7-4 can be derived in a formal way. Think about a person who makes predictions about N objects, such as tennis players or countries, and who recognizes n of these. The number n can range between 0 and N. The task is to predict which of two objects has the higher value on a criterion, such as which player will win the match. There are three possibilities: a person recognizes one of the two objects, none, or both. In the first case, one can use the recognition heuristic, in the second, one has to guess, and in the third, one has to rely on knowledge. The numbers nur, nuu, and nrr specify how often these cases occur (u = unrecognized, r = recognized). If one relies on the recognition heuristic and every object is paired with every other object once, we then get the following:
Number of correct predictions = nur© + nuu1/2 + nrr®.
The first term on the right side of the equation accounts for the correct inferences made by the recognition heuristic; for instance, if there are ten cases where one of the two objects is recognized and the recognition validity © is .80, then one can expect eight correct answers. The second term is for guessing, and the third term equals the number of correct inferences made when knowledge beyond recognition is used (® is the knowledge validity). In general, assuming that a person uses the recognition heuristic and © and ® are constant, it can be proven that the recognition heuristic will yield a less-is-more effect if © > ® (Goldstein and Gigerenzer, 2002).

11. Goldstein and Gigerenzer, 2002.

12. Schooler and Hertwig, 2005.

13. Gigone and Hastie, 1997. The majority rule has been reported for situations in which the correct answer cannot be proved by a group member (as it would be in the case of a math problem).

14. Reimer and Katsikopoulos, 2004. The authors show that the less-is-more effect in groups (reported in the next passage) can be formally derived in the same way, as illustrated by Figure 7-4.

15. Toscani, 1997.

16. Hoyer and Brown, 1990.

17. Allison and Uhl, 1964.

18. Oppenheimer, 2003. See also Pohl, 2006.

19. Volz et al., 2006. In this study, judgments followed the recognition heuristic in 84 percent of the cases, similar to earlier experiments. Moreover, when participants were partially ignorant, that is, they had heard of only one of the two cities, they got more correct answers than when they had heard of both cities.

20. Interview with Simon Rattle (Peitz, 2003).

Chapter 8: One Good Reason Is Enough

1. For experimental evidence that people often rely on only one or a few reasons, see Shepard, 1967, Ford et al., 1989, Shanteau, 1992, Bröder, 2003, Bröder and Schiffer, 2003, and Rieskamp and Hoffrage, 1999.

2. Schlosser, 2002, 50.

3. Dawkins, 1989, 158-61.

4. Cronin, 1991.

5. Gadagkar, 2003.

6. Grafen, 1990, showed that the handicap principle can work both for the evolution of honest signals and in the context of sexual selection.

7. Petrie and Halliday, 1994. The number of eyespots could in turn be inferred from the symmetry of the train; see Gadagkar, 2003.

8. See Sniderman and Theriault, 2004.

9. Menard, 2004.

10. Cited in Neuman, 1986, 174.

11. Also cited in Neuman, 1986, 132.

12. Sniderman, 2000.

13. The string heuristic is a realization of Coombs’s (1964) unfolding theory and the concept of proximity voting.

14. Gigerenzer, 1982. This study analyzed voters’ reactions to two new parties, the Greens and the European Workers Party (EAP, not reported here). Voters knew very little about the program of the EAP, but their preferences and judgments were as consistent as for the Greens, about which they knew more. There is an important methodological lesson in this (see Gigerenzer, 1982). If I were not interested in how the mind works, I would not ask what cognitive process underlies voters’ intuitions, but instead directly take their judgments and analyze them with a standard statistical package. Continuing, I would find and report that the correlations between Left-Right, preferences, and ecological ratings are mostly close to zero. Thus, I would erroneously conclude that preference orders and other issues cannot be explained by Left-Right, and that voters therefore have a differentiated system of reasons in their minds. I might never notice this blunder, cherishing the principle “let’s correlate first, and explain later.” There are patterns in intuition even when there are none in correlations.

15. Neuman, 1986.

16. Sniderman et al., 1991, 94.

17. Scott, 2002.

18. Bröder, 2000, 2003; Bröder and Schiffer, 2003; Newell et al., 2003. As in the study with parents, these experiments reveal individual differences; that is, people differ on the rules of thumb they use.

19. Keeney and Raiffa, 1993.

20. Tversky and Kahneman, 1982; on conservatism see Edwards, 1968.

21. Todorov, 2003. Bayes’s rule is named after the Reverend Thomas Bayes, to whom this rule is attributed. It computes the so-called likelihood ratios for the various possible differences in halftime scores because larger differences should have more weight than smaller ones. Take the Best, by contrast, just looks at who is ahead, and ignores by how much. Although some think of Bayes’s rule as the rational way to make decisions in the real world, it is actually impossible to follow this rule in problems of sufficient complexity because it becomes computationally intractable. Bayes’s rule can be used when only a few cues are known, but the complex rule is of little use for complex problems.

22. Gröschner and Raab, 2006. In a second study, 208 experts and laypeople were asked to predict the 2002 soccer world champion. Laypeople did significantly better, predicting the winner twice as often as experts. The laypeople more often went with the intuition that the team who had won the most championships beforehand would most likely win again (Brazil), and they were right.

23. When I first reported in a talk that one good reason was better than Franklin’s rule, a renowned decision researcher got up and said, “If you want to impress me, you need to show that one good reason can stand up to multiple regression.” We took up his challenge and showed for the first time that Take the Best can also outperform it (Gigerenzer and Goldstein, 1996, 1999). We published the data so that everyone could rerun the tests and check the claim. Many who could not believe their eyes confirmed the original result. The next objection was that we had demonstrated this only once. So we extended the test to a total of twenty real-world problems from psychology, economics, biology, sociology, health, and other areas. Multiple regression made an average of 68 percent correct predictions and Take the Best 71 percent. When the news spread in the scientific community, and others independently confirmed the results, the defenders of more-is-always-better suspected the fault no longer in the problems we used, but in how we used multiple regression. Some experts said that we should have calculated different versions of this method. We did, and found basically the same result (Czerlinski et al., 1999; Martignon and Laskey, 1999). Finally, we were able to prove some of the conditions under which the complex strategy cannot do better than Take the Best (Martignon and Hoffrage, 1999 2002; Katsikopoulos and Martignon, 2006). That put an end to this objection, but not to the debate. Suddenly multiple regression was no longer the issue; the argument now was that Take the Best needed to be compared with highly complex information-greedy algorithms from artificial intelligence and machine learning. We took up the challenge and found that in many situations, one good reason can predict more accurately than these extremely complex strategies (Brighton, 2006). The complex strategies tested included (1) connectionist models: feed-forward neural networks trained with the back-propagation algorithm, (2) two classic decision tree induction algorithms: classification and regression trees (CART) and C4.5, and (3) exemplar models: the basic nearest neighbor classifier and an elaborate model based on the Nosofsky’s GCM model.

24. This is a verbal summary of some of the analytical and simulation results reported in Gigerenzer, Todd, et al., 1999, Katsikopoulos and Martignon, 2006, Martignon and Hoffrage, 2002, Hogarth and Karelaia, 2005a, b, 2006. The crossing of the two lines in Figure 5-2 illustrates the problem of overfitting. One can define overfitting in the following way. Consider two random samples from a population (such as two years of temperature measures); the first year is the learning set and the following year is the test set. A model overfits the learning set if an alternative model exists that is less accurate on the learning set but more accurate on the test set.

25. The references for the following examples are in Hutchinson and Gigerenzer, 2005.

26. The original Roman calendar had ten months and the year began with Martius (March); Januarius and Februarius were added later. Julius Caesar changed the start of the year to the first of January, and Quintilis was renamed Julius in his honor; Sextilis later became Augustus in honor of Caesar Augustus (Ifrah, 2000, 7).

Chapter 9: Less Is More in Health Care

1. Naylor, 2001.

2. Berg, Biele, and Gigerenzer, 2007. Informative medical literature, such as the Guide to Clinical Preventive Services by the U.S. Preventive Services Task Force, 2002a, is easy to find in a university library or online.

3. Merenstein, 2004.

4. Ransohoff et al., 2002.

5. U.S. Preventive Services Task Force, 2002b.

6. Lapp, 2005.

7. Etzioni et al., 2002.

8. Schwartz et al., 2004.

9. U.S. Food and Drug Administration; see Schwartz et al., 2004.

10. Lee and Brennan, 2002.

11. Gigerenzer, 2002, 93.

12. Domenighetti et al., 1993. Note that Switzerland removed financial barriers to medical care for the whole population almost one century ago. Thus, the rate of treatment (including overtreatment) in the general population is not distorted by uninsured citizens having no access to treatment, as in countries without universal health care (this may explain why a study in the United States did not find different hysterectomy rates for doctors’ wives; see Bunker and Brown, 1974).

13. Deveugele et al., 2002; Langewitz, et al. 2002.

14. Kaiser et al., 2004.

15. Wennberg and Wennberg, 1999.

16. Wennberg and Wennberg, 1999, 4.

17. Elwyn et al., 2001.

18. See the reader by Dowie and Elstein, 1988.

19. Elwyn et al., 2001. Concerning our program to improve physicians’ and patients’ intuitions about risk and uncertainties, see Gigerenzer, 2002, and Hoffrage et al., 2000.

20. Pozen et al., 1984.

21. See Green and Yates, 1995.

22. Green and Mehr, 1997.

23. Corey and Merenstein, 1987; Pearson et al., 1994.

24. For details, see Martignon et al., 2003.

25. Recall that when using the heart disease predictive instrument, the physician calculates a number for each patient, then compares it to a threshold. If the number is higher than the threshold, the patient is sent into the care unit. This threshold can be set high or low. If it is set high, then fewer patients will be sent into the care unit, and more misses will result. This corresponds to the squares at the left of Figure 9-3. If the threshold is set low, then more people will be sent into the unit, which increases the number of false alarms, as shown by the squares at the right.

26. In a replication study in two other Michigan hospitals, which used the successor of the heart disease predictive instrument, the Acute Coronary Ischemic Time-Insensitive Predictive Instrument (ACI-TIPI), a fast and frugal tree again did as well as the complex method (Green, 1996).

Chapter 10: Moral Behavior

1. Browning, 1998, xvii. I chose this sensitive example because it is one of the best-documented mass murders in history, with the unique feature that the policemen were given the opportunity not to participate in the killing. If you know of other similar examples, please let me know. My short account cannot do justice to the complexity of the situation, and I recommend consulting Browning’s book, including the afterword, in which he deals with his critics such as Daniel Goldhagen. Browning(e.g., 209-16) offers a multilayered portrayal of the battalion during their first and subsequent mass killings. The largest group of policemen ended up doing whatever they were asked to, avoiding the risk of confronting authority or appearing to be cowards, yet not volunteering to kill. Increasingly numbed by the violence, they did not think what they were doing was immoral because it was sanctioned by authority. In fact, most did not try to think at all. A second group of “eager” killers who celebrated their murderous deeds increased in numbers over time. The smallest group was the nonshooters, who, with the exception of one lieutenant, however, neither protested against the regime nor reproached their comrades.

2. Browning, 1998, 71.

3. Johnson and Goldstein, 2003. Note that this figure is the proportion of citizens who are potential donors by law, not the actual donation rate. The latter depends on how well the process that matches a donor with a recipient is coordinated and how well the staff is trained—including how fast the donor, typically a traffic accident or stroke victim, is transported to a hospital. In the years 1996 through 2002, by far the best organization and highest true donation rate was reached in Spain, a country with a presumed consent policy; that is, people are potential donors by default.

4. Johnson and Goldstein, 2003.

5. Johnson et al., 1993.

6. Haidt and Graham, in press, base their five moral dimensions on the work of Shweder et al. (1997), where harm and reciprocity concern the ethics of autonomy, hierarchy and ingroup the ethics of community, and purity the ethics of divinity. The connection of these five dimensions with individual, family, and community (rather than with autonomy, community, and divinity) is not of their doing but my own responsibility. See also Gigerenzer, in press.

7. Kohlberg et al., 1983, 75. In this article, the authors reformulated Kohlberg’s (1981) original theory. The following evaluation of the evidence is based on Shweder et al., 1997.

8. Haidt, 2001.

9. Harrison, 1967, 72.

10. Haidt, 2001, 814; see also Nisbett and Wilson, 1977, and Tetlock, 2003.

11. Laland, 2001.

12. Terkel, 1997, 164.

13. The Bail Act 1976 and its subsequent revisions; see Dhami and Ayton, 2001.

14. Dhami and Ayton, 2001, 163. The following quote is from Dhami (August 2003, personal communication).

15. Dhami, 2003.

16. Dhami and Ayton, 2001.

17. Gazzaniga, 1985.

18. Numerous versions of consequentialism exist; see Williams, 1973, and Downie, 1991. Sunstein, 2005, provides an interesting discussion of rules of thumb and consequentialism.

19. Daston, 1988.

20. Bentham, 1789/1907; see Smart, 1967. The hedonic calculus is from Bentham’s chapter 4. As requested in his will, Bentham’s body was preserved and exhibited in a wooden cabinet at University College London, where, topped with a wax head, it can still be viewed today.

21. Dennett, 1988.

22. Sunstein, 2005; Viscusi, 2000.

Chapter 11: Social Instincts

1. Humphrey, 1976/1988, 19; see also Kummer et al., 1997.

2. Richerson and Boyd, 2005.

3. Cronin, 1991.

4. Darwin, 1874, 178-79.

5. Sober and Wilson, 1998.

6. See Cosmides and Tooby, 1992; Gigerenzer and Hug, 1992.

7. Frevert, 2003.

8. Resche, 2004, 723, 741.

9. Mervyn King, “Reforming the international financial system: The middle way.” Speech delivered to a session of the money marketers at the Federal Reserve Bank of New York, September 9, 1999.

10. Gallup International, 2002.

11. The neurologist Antonio Damasio, 1994, 193-94, reported a patient called Elliot with a damaged frontal lobe. One day, Damasio asked him when the next session should take place and suggested two alternative dates, just a few days apart from each other: “For the better part of a half hour, the patient enumerated reasons for and against each of the two dates: Previous engagements, proximity to other engagements, possible meteorological conditions, virtually anything that one could reasonably think about concerning a simple date…. He was walking us through a tiresome cost-benefit analysis, and endless outlining and fruitless comparison of options and possible consequences.” When Damasio advised him to take the second date, Elliot simply said, “That’s fine.”

12. Richerson and Boyd, 2005.

13. For an evolutionary theory of social change, see Boyd and Richerson, 2005.

14. Lightfoot, 2003.

15. Hertle, 1996, 7, 245. The following account is based on Hertle’s research.

16. Hertle and Stephan, 1997, 42.