B Heuristics, Biases, and the Hypothesis

Scientific thinking is often held up as the standard—not exactly a golden one, but an accepted standard—for rational, objective thinking and reasoning.¹ The insights we gain from studying scientific thinking are supposed to find their way into daily life as aids to practical thinking.

Bias has become a big concern because of the Reproducibility Crisis; however, it has been at the center of a heated debate in cognitive psychology for a long time. This debate matters to those of us interested in the scientific hypothesis be­cause, as we've already seen in Chapter 10, some critics detect an unsavory con­nection between the hypothesis and bad forms of bias. We begin with a look into what it means to be biased.

Bias is a complex topic; psychologists distinguish well over 100 different forms of cognitive bias.³ We'll group the great majority into a few broad categories be­fore zeroing in on a few. There is a lot of angst about biases that we can call social biases that lead to malicious behavior toward others; that isolate, denigrate, or discriminate against them; or that cause scientific misconduct.

Most of the cognitive biases that we'll be concerned with are universal traits of human thinking that frequently result from using mental shortcuts called heuristics. Heuristics are normally useful and readily-available tools for simpli­fying complex tasks; they come to us so naturally that they're easy to overlook until they create problems. Heuristics and biases arise automatically and uncon­sciously; we'll go over a few examples shortly.

There are two main schools of thought regarding heuristics, and they are dis­tinguished principally by how heavily they weigh the useful versus the harmful, the rational versus the irrational sides of heuristics. One school is founded on insights derived from evolutionary psychology, the discipline that holds that the foundational structures and processes of our minds are adapted to succeed in the kinds of environment that our ancient apelike ancestors encountered. This school is associated with Gerd Gigerenzer,⁴ and it focuses on the advantageous nature of heuristics and biases—their ecological rationality. Gigerenzer sees heuristics as “efficient cognitive processes that ignore information”⁵ so that we can act quickly and effectively. This school is particularly impressed by cases in which heuristics outperform purely logical problem-solving strategies. Heuristics make sense be­cause they help us navigate the world safely and productively and were, there­fore, selected for by evolution.

The other school of thought regarding heuristics, personified by Nobel Prize­winning psychologist Daniel Kahneman,⁶ thinks they are “simple procedures[s]” that help “find adequate, though often imperfect, answers to difficult questions.”⁷ Despite acknowledging its potential benefits, Kahneman's school devotes its attention to the disadvantages of heuristic thinking; to the cognitive errors we commit and the biases we fall prey to when using heuristics.

These opposing points of view have important, though widely divergent, implications for scientific thinking, and, in order to compare them, we need to expand on their interpretations of rationality. We'll begin with Gigerenzer's ideas.

12. B.1 The Fast and Frugal Program: Heuristics as

Useful Mental Tools

For Gigerenzer, the epitome of a perfectly designed heuristic is the gaze heu­ristic that makes it possible for a baseball outfielder to catch a fly ball or a dog to catch a Frisbee. While you could, in principle, calculate the flights of baseballs and Frisbees, even if you knew all of the relevant variables—the object's initial speed and launch angle, wind direction and velocity, etc.—and you never do, it wouldn't do you any good. Before you finished the calculations, the thing you're trying to catch would already have hit the ground.

Yet you and your dog can perform flawlessly by following a simple heu­ristic: if the object is already high in the air, look right at it, and run toward it just fast enough that your eyes remain fixed on it at the same gaze angle. Eventually your path will intersect the object's and, if you don't fumble, you'll make the catch. Modify the heuristic slightly, and it will work while the ob­ject is still rising. The gaze heuristic represents an ideal of the fast-and-frugal solutions that, for countless millennia, simplified the otherwise overwhelming

complexity of the world for us, the species that Gigerenzer calls “homo heuristicus.”⁸

Not all heuristics have been programmed by evolution.

Though few heuristics are as elegant as the gaze heuristic or as narrowly fo­cused as the 1/N heuristic, most help us do better, most of the time, than we would without them. Our reliance on heuristics is, therefore, rational, provided that we keep in mind the words of Nobel Prize-winner Herbert Simon, that we have only “bounded rationality”¹⁰ to begin with. Our mental capacities are lim­ited and, hence, perfect, logical rationality is rarely even an option.

Gigerenzer's program also relies heavily on another of Simon's princi­ples: “satisficing,” a Scottish coinage that captures the spirit of “satisfying” and “sufficing.” Satisficing is the judicious middle ground. When you are satisficing in making a choice, you are not trying to maximize an outcome; instead, you are weighing your alternatives with a minimal standard, an “aspiration level” that you're trying to achieve, and, as soon as you discover one, you take it without further ado or analysis. Take-the-best¹¹ is an example of a “one-reason” heuristic that we resort to on a daily basis. More elaborate fast-and-frugal heuristics are constructed from basic principles like these, and I'll have more to say about them later. Both bounded rationality and satisficing are on display in the resolution to the bias-variance dilemma that accompanies many scientific conundrums.

12. B.1.a The Bias-Variance Dilemma

One of Gigerenzer's pivotal insights is that “less is more,” referring to the capa­bility of stripped-down heuristic solutions to surpass more complicated ones in predictive accuracy. He accounts for this by pointing to the bias-variance di­lemma, also sometimes called the bias-variance tradeoff, which he uses to illus­trate how bias resulting from the use of heuristics can be a good thing.¹²

The dilemma comes from the fact that the true relationship between what you observe and the aspect of reality that you want to know about is obscured by un­known amounts and kinds of noise—uncontrolled variability in your data. (The variability has its origins in two sources: true individual differences, which causes sampling error, and irreducible error, which is the random error in our measuring instruments, the environment, etc. Irreducible error affects all measurements, so we rarely give it special attention.) When you’re trying to understand the world in terms of hypotheses or models, you have to decide how to deal with variability and still portray reality in meaningful ways.

12. B.1.b The Bias-Variance Dilemma: Example

Say you’re interested in predicting young children’s heights as they age be­tween 4 and 10 years old. With permission, you go to your local public school and measure the heights of 100 preschool kids. When you plot their heights in inches against their ages in months, the general trend is obvious—they get taller as they get older—and there is much variability; at any given age, some children are shorter and some taller than others. What’s the best way to capture the overall relationship between height and age?

One way would be to connect all the dots in the graph, beginning with the height of the youngest child, drawing a line to the height of the next oldest, to the next, etc., finishing at the height of the oldest. The jagged line would accurately represent the growth rate of your group but it would have no predictive value; if you measured a different group of kids of the same ages, the same jagged line would not connect the dots in the second group.

Alternatively, you could mathematically fit a straight line to your data and use the line to encapsulate your conclusions about children’s growth. The fitted line would depict the drift of the growth and simplify the plot enormously, although it would miss all of the variable information about their individual heights. The straight line would be both a crude summary of existing data and an implicit hypothesis. You’d probably feel that, if you measured the heights of other kids in the same age range, the straight line would be a fair approximation of their growth rate as well. Most importantly, because you decided to fit a line, rather than another function, the line would be a form of bias. With it, you imposed your opinion on the dots; there was no line in the data. Bias, in this general sense, means weighting information unequally—diminishing some parts of it while paying a great deal of attention to other parts. Bias may be good or bad; reason­able or unreasonable. Bias reveals an inclination toward a particular outcome; it is not a detailed defense of the outcome. This is the tradeoff: increased robust­ness and predictive cogency against the loss of detail. When we use heuristics, we introduce bias.

Automatic, implicit hypothesis generation represents a type of cognitive heu­ristic thinking that is usually adaptively accurate, like the generally trustworthy guidance that we get from our sensory systems. Because of them, we confidently act as if the external world is stable, predictable, and benign. Our countless im­plicit hypotheses about it are rarely falsified even though they’re tested all the time; that's why we get away with casually assuming that they're accurate and are taken aback from time to time when we find that they're not.

Still, we shouldn't forget that biases can also throw us off course. Like sen­sory physiologists who study visual illusions to find out how our visual systems normally operate, cognitive scientists study biases to discover how our cognitive systems normally operate.

12. B.2 Heuristics as the Source of Bad Cognitive Bias

In their Prospect 'Theory of human decision-making under conditions of uncer­tainty, Kahneman and Amos Tversky¹³ detailed what they saw as the “irration­ality” of typical human thought, where rationality for them refers to the way in which an omniscient, infallibly logical observer would think. They concluded that, when we use heuristics, the biases we adopt are errors in thinking: they should be stamped out, although Kahneman is pessimistic that psychology can teach us to do that.

In Thinking, Fast and Slow,¹⁴ Kahneman puts forward a two-system interpre­tation of heuristics and biases. His vision is that each of us has two fundamen­tally different cognitive systems—System 1 and System 2—that don't follow the same rules. Roughly speaking, System 1 is at work all of the time processing in­formation automatically in a swift and effortless, though often careless, manner. System 1 gives rise to our implicit hypotheses, heuristics, and biases. System 2 is plodding, logical, and careful, albeit often lazy and hard to engage. When we rig­orously examine and test a hypothesis, we're using System 2. Although they can be associated to some extent with separate brain regions, the Systems are mainly convenient labels that designate markedly dissimilar forms of cognition.

Kahneman accounts for many biases as the result of cognitive illusions. What are cognitive illusions? They're a bit like the sensory illusions that we discussed in Chapter 11. You experience an illusion when your brain, in automatic- processing mode, presents you with a discrepancy (e.g., you see two lines of unequal lengths, but your ruler measures two identical lines). Analogous discrepancies occur be­tween our gut-level, instinctive responses to mental challenges and the solutions that we get from dispassionate, methodical analysis. Cognitive illusions can be as seductive as the sensory kind. Here is a quick example: If a ball and a bat to­gether cost $5.50, and the bat costs $5.00 more than the ball, how much does the ball cost? If you reflexively answered, “$0.50,” you were the victim of a cognitive illusion. Think about it.¹⁵

Kahneman and colleagues see cognitive illusions as evidence of irrational be­havior. This is the point where we need an index of rationality, a standard like a ruler, so we can judge it objectively. If you're accustomed to thinking about ra­tionality as a fuzzy concept with no place outside of philosophy class, you may be surprised to learn that a real-world standard exists; it is called Expected Utility 'Theory (or just Utility Theory).¹⁶ (If you're unfamiliar with the theory, take a mi­nute to go over Box 12.1.)

12. B.2.a What Does Rational Decision-Making Look Like?

Economists have long been obsessed with how people make financial decisions, and lessons about decision-making characteristically begin with examples involving money. Money problems are ubiquitous because everybody can readily relate to them, and, furthermore, these decisions have quantifiable outcomes so we can identify good and bad decisions unambiguously. Later we'll see how to extend Utility Theory beyond the wallet.

Expected Utility Theory says that the psychological value of money is de­termined solely by its utility—the amount of personal satisfaction, measured in goods and services, that you can buy with it. The utility of $1,000 is deter­mined by how much desirable stuff you can get for $1,000. Rational decision­makers base their decisions on how much utility they expect to have as a result of their decisions, and they always try to maximize the absolute amount of their expected utility (e.g., to have the most money). Nothing else matters. And ra­tional decision-makers always act selfishly to maximize their utility; they pay no attention to the needs or wants of others, the benefits to society, etc. (Rational decision-makers don't have many friends.)

As I mentioned, it is standard practice to discuss Utility Theory in terms of money and fiscally related “utilities,” however the theory “serves as a model for various psychological processes, including motivation, moral sense, attitudes, and decision-making”¹⁷ In other words, you can evaluate the rationality of all sorts of behavior by looking at whether or not it maximizes the psychological “utility” that's involved and otherwise adheres to the precepts of Utility Theory. That's more or less what you're doing when you weigh the “pros” and “cons” of spending time with a friend who, on the one hand, is a lot of fun to be with yet who, on the other hand, can be a real jerk. Ultimately, you opt for the choice that you expect to make you happiest, or you should if you want to be rational by the lights of Utility Theory.

Illusions, sensory or cognitive, are compelling because they depend on low- level, reflexive information processing; they affect us before we know it. When you put your cold-adapted and warm-adapted hands into the same bucket of lukewarm water, you felt the water temperature differently with each hand and knowing the truth—that the temperature had to be the same for both—didn't change how it felt. Breaking the grip of a cognitive illusion like the bat-and-ball problem can be almost equally hard to do.

Cognitive illusions resemble sensory illusions in ways besides their almost irresistible nature, however. Kahneman and Tversky postulate that we evaluate gains and losses with respect to a given reference point, not in absolute terms.

Box 12.1 Expected Utility Theory: A Brief Overview

For the basics of the theory, we need to start with the related notion of ex­pected value, which is a statistical term. You can think of expected value as the average value of something that you don't have (or of something that hasn't happened), and it depends on the probability of getting that thing (or having it happen). To calculate expected value, you multiply probability times value. For instance, if someone offered you a lottery ticket that had a 50% chance of winning $100, you'd multiply the probability of the win, 0.5, times its value, $100, and find that the expected value of the ticket is $50.

Utility is a term from economics that rates how you, personally, value something; utility is the amount of satisfaction that you get from a particular good or service. Utility is, therefore, subjective value. It is correlated with the value of money, but is not equivalent to dollar value because two people might get different amounts of satisfaction from a certain good or service. If you're a vegan, the utility of a meal at an upscale vegan restaurant could be high even if the meal were a bit expensive; if you're a committed carnivore, even a cheap price would probably not increase the utility of a vegan meal for you. The con­cept of utility even applies to money itself: if you're broke and your rent is due today, you might find a “payday lender's” exorbitant interest rate acceptable, despite the fact that it will cost you more money in the long run. Like expected value, expected utility is an average. Expected utility depends on what some­thing would cost, how much you want it, and the chance that you'll get it.

Modern expected utility theory, often referred to as rational economic theory, was formulated by John Von Neuman and Oskar Morgenstern, (see Note 16) and includes a set of logical axioms that define rational decision­making. For example, if you prefer option A over option B, and option B over option C, then, according to the theory you must prefer option A over op­tion C. If you chose apples over pears, and pears over bananas, then you must prefer apples over bananas. If your behavior conforms to the axioms, you are acting rationally, otherwise not.

The only thing that matters for economically rational thinkers is the end point; their happiness depends only on the total utility of what they have, not whether they gained or lost in getting that amount. Expected utility theory does recognize that the utility of money does not increase linearly with how much you have; someone whose financial worth is only $1,000 would value a windfall of $100 more than would a millionaire. The relationship, according to Daniel Bernoulli, one of the historical forbears of the modern theory, is logarithmic: the millionaire should value $100,000 roughly as much as the less-well-off person would value $100.

Imagine two people, Joe and Jane. initially Joe has $1,000 and Jane has $3,000. One day, Joe inherits $1,000 from a rich uncle and Jane loses $1,000 in the stock market. Now they both have the same amount of money, $2,000 and, by the rational standard of Expected Utility Theory, both should be equally happy. Clearly, this is ridiculous; everybody knows that Joe will be ecstatic, and Jane will be distraught. Real people care not just about where they are financially, but how they got there. Like the hands in the lukewarm water, Joe and Jane experi­enced opposite emotions when they were exposed to their new common level of wealth. They were either happy or sad depending on how their fortunes had changed from their individual reference levels, not what their final absolute level was. And this, in turn, suggests that they had “adapted” to their preexisting ref­erence levels of wealth. Rational economic agents don't experience any of these emotions.

And rational agents respond to losses and gains in a symmetrical manner: the prospect of losing $100 is just as unpleasant as the prospect of gaining $100 is pleasant. Do you agree? Would you take a bet on the toss of a fair coin if you were to win $100 if it came up heads, but lose $100 if it came up tails? (If you would normally don't take bets of any kind, please relax for a moment and consider this one.) No? How about if you were offered $150 for the win against $100 for the loss? Still no? Expected Utility Theory says that you definitely should take the bet. The expected value of a gain of $150 is 0.5 x $150 = $75, and the expected value of a loss is 0.5 x -$100 = -$50. The overall expected value of the bet is the sum of the two possible outcomes: $75 - $50 = +$25 (i.e., a net gain, meaning that on average you'll come out $25 dollars ahead if you take bets like this). Most people don't take it. Theory be damned, we want a chance to win at least $200 before we're willing to risk losing $100. In our account books, losses “count” for much more than gains. In Prospect Theory, Kahneman and Tversky called this asymmetry loss aversion. According to the theory, loss aversion is a powerful psy­chological bias that comes into a large variety of decision-making processes.

12.