You have earned a badge! Click me to read more.Click on any reward badge in the footer to review take home messages of this or other elements.
Taming Uncertainty comes with 17 interactive elements that highlight concepts from the book. Each time you complete a new interactive element, you will receive a badge that contains the key message you encountered. Here you can view all the badges you have earned and their key messages. Complete them all and become a Decision Master!
Right now you have earned 0 of 17 badges ().
In this element, our aim was to help you give you first-hand experience on how the different decision strategies perform when they must reach a decision. We show in the chapter that under uncertainty some heuristics—surprisingly those that ignore probabilities—perform almost as well (and sometimes even better) than those that use a maximization calculus.
We also hope you saw that simulation can be a wonderful tool to explore complex problems and to test and challenge intuitions about them. Varying elements of simulations, one can gain insight into the conditions (environments, sample sizes, problem types,...) under which observed relationships are observable and thus learn about the robustness or fragility of results. Often simulations can supplement both mathematical analyses and empirical studies.
The incidental learning task is a good model of how people pick up statistical structures from the environment. We rarely have the luxury of obtaining feedback, and we often learn while our attention is focused on something else: Making good choices. Learning risk–reward structures helps people use the environment to make decisions in the many situations where probabilities are not explicitly stated. By learning the risk–reward structure of an environment, people can exploit previously learned structures to infer the probability directly from the payoff. This strategy can be useful in a wide range of situations. For more information, please see Chapter 3.
In many critical aspects of everyday life, no objective social statistics are available. Taking a mental poll of the people one knows personally is an effective strategy for reducing this uncertainty and forming approximate representations of the world’s social texture. Natural structures in a person’s social network can guide and limit this search for relevant information in social memory.
The goal of this task was to show you that it is not inevitable that limited search for instances in memory comes at the price of very low accuracy. You saw that, under conditions that are arguably common in natural environments—when the frequency distribution of the events is skewed and the events occur in spatial clusters—heuristic sampling that starts with a very small sample and sequentially expands search only if that sample does not yield sufficient evidence represents a useful tool for judging event frequencies. Because conditional and ordered sampling potentially ignores redundant information in later social circles, it can achieve higher accuracy than can unconditional random draws of samples of the same size.
Playing against opponents using different decision policies reveals the strong dependence of your payoffs on how an opponent plays. Consequently, facing strategic uncertainty–not knowing what decision policy your opponent is using–significantly complicates the game and your decision regarding how to play.
In strategic decision-making, the performance of decision policies depends not only on their interaction with the environment (such as the types of games played, the degree of conflict generated by said games, missing payoff information etc), but also on the distribution of the opponent decision policies in the population. By varying this distribution using this tool, you investigated the robustness of decision policies to strategic uncertainty. In particular, the L1 and D1 heuristics remain highly competitive for a wide range of possible distributions of opponents. By contrast, decision policies that make specific assumptions about their opponent, such as the Nash equilibrium and MaxMax, performed well only when these assumptions matched the characteristics of the majority of opponents. Otherwise, they performed relatively poorly, that is, they were not robust to strategic uncertainty.
The modern food environment is replete with uncertainty. One source of uncertainty is not knowing how much sugar is in our food, even though sugar consumption is a potential contributor to overweight and obesity. The World Health Organization recommends reducing consumption of free sugars to less than 10% of total daily energy intake (50 g, or 16 sugar cubes’ worth, for an average adult). People need a good understanding of how much sugar is in the foods they eat in order to be able to meet those guidelines. The goal of this task was to help you gauge how well you estimate how much sugar is in some of the foods you may eat, and to show you how well your fellow readers perform in the same task. Dallacker, Hertwig, and Mata (2018) found that parents considerably underestimated the sugar content of most foods and beverages—for instance, 92% of parents underestimated the sugar content of yogurt by, on average, 7 sugar cubes. The degree to which they underestimated sugar content was associated with a higher risk of their child being overweight or obese. For ideas about how parents can use this information to help their children eat better see Chapter 6 in Taming Uncertainty.
You just experienced the experimental conditions typically used to study the description–experience gap. In decisions from description, the canonical finding is that people make choices as if they overweight the rare event. So in problem 1 they prefer the safe option and in problem 2 they prefer the risky option; when these same gambles represent losses (problems 3 and 4) the preferences reverse and they prefer the risky option in problem 3 and the safe option in problem 4. This has been called the fourfold risk pattern.
Here’s where things get even more interesting: The full pattern typically reverses in decisions from experience, where people make choices as if they underweight the rare event.
How did you choose? How does this stack up against other respondents? For more information on the fourfold risk pattern and the role that search and exploration play in it please refer to Chapter 7 of Taming Uncertainty.
By simulating the exploration and decision making parameters of hypothetical decision makers and comparing them to real data, you were able to discover the model that best describes behavior in decisions from experience. If you have tried long enough, you should have observed that a per-option sample size of 10, minimal recency and noise, and, importantly, alpha, lambda, and gamma equal to 1 match human behavior very closely. This implies that for problems involving a risky and safe option, the kind typically studied, individuals appeared to have weighted outcomes and probabilities in a linear fashion, contrasting the elaborate models needed for decisions from description.
This element demonstrated the different effects uncertainty can have on the intertemporal choices people make. In this case, the beliefs people have on when a payoff will be delivered can and should influence the degree of patience people exhibit when making intertemporal choices.
In this interactive element we have shown you the experimental materials that we used to study investment decisions. In those experiments, we found that investors who learned from experience made different decisions than those who learned from a graph. We also gave you the chance to play the investment game in different markets. If you chose to play the game, you may have used one of the several heuristics that we identified in our study. Which you did you use? We observed in out studies that the most frequent heuristic to navigate financial uncertainty was “momentum trading”, where people increase their risk exposure following a market rise, and reduce it following a drop.
If you chose to play the game under both conditions, you may have noticed that when learning about a shock from experience, you were inclined to take less risk than when learning from a graph. This phenomenon is called the “depression babies effect”, and we observe it our experimental studies.
In the next interactive element, we show you how simulated experience can be harnessed to help people make better financial decisions.
In this interactive element we have shown you the Risk Tool and its components as addressed in chapter 10 and originally proposed by Kaufmann, Weber, & Haisley (2013). Each component increases the degree of experience that you get. In the description component, you only saw your allocation and its likely outcomes in text. In the experience component, you were able to sample from any proposed allocation. In the distribution component, you were able to see the whole possible distribution of outcomes of your proposed allocation. And finally, in the Risk Tool, you were able to simulate, either slowly or quickly, how the distribution of outcomes realizes. Which one did you prefer? Which component was most helpful to make an investment decision?
Here we illustrate how combining estimates of individuals into a single crowd estimate typically outperforms the average individual when answering general knowledge questions. The crowd estimate is defined as the median value across all individual estimates. Accuracy is defined as the absolute distance between an estimate and the true value (also known as “absolute error”); this implies that smaller values indicate higher accuracy.
Depending on your knowledge about a specific question, the crowd estimate is either more or less accurate than your personal estimate. But irrespective of whether or not you have beaten the crowd on a particular question, note that the error of the crowd estimate is typically smaller than the error of the average crowd member (“wisdom of crowds”). This is because errors in opposite directions (i.e., underestimations vs. overestimations) cancel out when aggregating independent estimates. See Box 13.2 for more details on why and when diverse errors cancel out.
The main idea conveyed by this interactive element is that simple decision rules, such as deciding randomly, following the minority, or imitating the most popular choice, can have unplanned collective consequences.
In the most-popular mode, for example, individuals tend to follow their neighbors. At the individual level, this can be an efficient strategy to deal with the uncertainty of the environment because one can benefit from the knowledge of others. However, if too many people adopt the same behavior, the best exit can become congested, creating a maladaptive collective pattern. The efficiency of a strategy is, therefore, a matter of circumstance.
In this interactive element, you saw how the processes of mutation, selection, and inheritance from generation to generation can produce agents that succeed at a task where their ancestors failed. These same processes allowed us to evolve our cognitive capacities, from memory to decision-making mechanisms. Since evolution in humans is almost impossible to observe in real time, computational models like the one running here allow us to study abstract properties of evolution in action. In turn, we can develop and inform theories of how real organisms process information based on observations from the evolutionary models.
In the chapter, we examine some of the ways computational evolutionary models have been used to deepen our understanding of how organisms grapple with uncertainty over generations through evolution.
You have earned a badge! Click me to read more.Click on any reward badge in the footer to review take home messages of this or other elements.