Reviewed by
Marta Posada
INSISOC University of Valladolid
The book is a collection of revised versions of Gigerenzer's most recent articles, examining the rationality of individuals not from the perspective of logic or probability, but from the point of view of the observed human behaviour and its interaction with the environment. The book deals with two complementary modes of thinking of the human mind: heuristic and statistical. Bounded rationality by heuristic thinking is the key to understand how real people make decisions when time and information are limited. Statistical thinking is the key to describe the limits of human inference under uncertainty. It covers issues dealt in previous books written by Gigerenzer (Gigerenzer et al. 1999; Gigerenzer 2000; Gigerenzer and Selten 2001) but updated with new references.
In the first six chapters the reader is introduced to heuristic thinking by the following four notions of human rationality:
(i) Unbounded rationality. Its goal is to explain decision making, assuming people behave as if they were omniscient (perfect knowledge) and omnipotent (unlimited computational power) in a world without uncertainty. Gigerenzer opposes this notion because this assumption has already been considered unrealistic by Herbert Simon forty years ago (e.g., Chapter 1).
(ii) Bounded rationality by optimization under constraints. Gigerenzer opposes this notion because he believes it to be a modification of the previous, but diminishes omniscient by introducing the need to search for information and resulting costs (e.g., Chapter 1).
(iii) Bounded rationality by the heuristics-and-biases. Its main goal is to understand the cognitive processes in decision making that produce both valid and invalid judgements. Gigerenzer opposes this notion because judgements deviated from the rational answer (called Cognitive Illusion according to Daniel Kahneman and Amos Tversky) cannot be attributed to some deficit in the human mind, but to logical norms. To support this statement, some updated examples of phenomena, which were first interpreted as cognitive illusions but later re-evaluated as reasonable judgments, have been reported in this book (e.g., Chapters 1 and 5). Following Herbert Simon's analogy, bounded rationality is like a pair of scissors: the mind is one blade and the environment is the other. The Heuristics-and-biases notion only studies the cognitive blade. However, one blade alone does not cut well.
(iv) Ecological rationality. Its goal is to relate human decision making to the environment in which humans operate. Examples have been used to demonstrate that people actually use fast (in little time) and frugal (with little information) heuristics to make good decisions instead of solving complex differential equations. Moreover, people sometimes ignore information. This is not a bad error but a good error (e.g., chapter 4). The study of heuristics involves the following items: (i) identifying the adaptive toolbox which contains a repertoire of fast and frugal heuristics (e.g., chapter 2) and its change over the life course; (ii) identifying the structure of environments that a given heuristic can exploit, that is, the kind of problems it can solve; (iii) designing heuristics and/or environments to improve decision making in applied fields such as healthcare, law and management (e.g., Chapter 6).
In the next five chapters the reader is introduced to statistical thinking, which can help to remedy problems such as statistical innumeracy and rituals in social sciences.
Statistical innumeracy is the inability to think with numbers that represent risks and uncertainties. Gigerenzer shows how people misunderstand what a probability means in predicting weather (e.g., Chapter 8) and in communicating clinical risks (e.g., Chapter 9). Gigerenzer believes that the misunderstanding of statistical information lies in the external representation of information and not in the human mind. To demonstrate this, three confusing representation of data in healthcare are examined: single-event probabilities (the probability that a woman has breast cancer is 0.8%), conditional probabilities (if the woman has breast cancer, the probability that she will have a positive result on mammography is 90%) and relative risks (undergoing mammography screening over the age of 50 reduces the risk of dying from breast cancer by 20%). The key to defeat statistical innumeracy is to use alternative representations of statistical data such as natural frequencies (the way humans have encountered statistical information): single-event probabilities (8 out of 1000 women, who undergo a mammography, have breast cancer), conditional probabilities (7 out of those 8 women had a positive result on mammography) and relative risks (10 women died from breast cancer out of 1000 women who did not undergo mammography screening, whereas 8 out of 1000 women who did undergo it).
Statistical thinking, whose historical evolution is described in this book (e.g., Chapter 10), has been largely eliminated by statistical rituals (that is when one, and the same, method is automatically used to solve every problem). Although the discussion is focused on experimental psychology research, it can be applied to social sciences in general (e.g., Chapters 7 and 11). Gigerenzer proposes a sampling taxonomy and encourages to reflect which method is most appropriate and in which circumstances (e.g., Chapter 7).
Heuristic and statistical thinking are linked in Chapter 12 by comparing children and adults' capacity to solve Bayesian problems. Results provide evidence that children can reason in Bayesian way if information is provided in natural frequencies. Although the set of heuristics strategies used by children and adults are analysed, a developmental trajectory in strategies is left for future research.
This book should be read by all scientists interested in understanding human behaviour on decision making, or in knowing the limits of human inference under uncertainty in order to provide statistical information in an understandable manner. This book is useful for JASSS community because in artificial models, statistical performance should be reported. In particular, it should appeal to those scientists interested in modelling real agents' decisions, but not omniscient-omnipotent agents.
ReferencesGIGERENZER , G, and Selten R (eds.) (2001) Bounded Rationality: The Adaptive Toolbox. Cambrige, MA: MIT Press
GIGERENZER, G, Tood PM, ABC Research Group (1999) Simple Heuristics That Make Us Smart. New York: Oxford University Press
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2009