How do people make decisions? Conventional, logical thinking would have us believe that we base our decisions on optimisation; that people follow a principle of probability and reason, so as to maximise their utility. While that may make for a "beautiful mathematical system", what happens when the assumptions of neoclassical economics fall short? A question posed by Gerd Gigerenzer, Director at the Center for Adaptive Behavior and Cognition (ABC), Max Planck Institute for Human Development, Berlin.
Speaking at a forum organised by SMU's Behavioural Sciences Institute, Gigerenzer told the crowded room of mostly academics this story: "A professor from Columbia University had an offer from Harvard. He couldn't make up his mind - whether he should accept or reject... So a colleague took him aside and said, 'What is your problem? Just maximise your expected utility! You always tell your students to do so.' Exasperated, the professor responded, 'C'mon, this is serious.'"
In an uncertain world, people require some biases in order to make better inferences, he said. 'Optimisation', as seen through rational economic frames, may not always be within the reach of the human mind. And this is where 'heuristics' may be of some help.
A widely-published psychology professor in the area of heuristics and bounded rationality in decision making, Gigerenzer explained that 'heuristics', put simply, is a rule that directs focus to areas that matter, and in the process, blocking out non-essential information. The end result: faster decisions with less information. Seems too good to be true, and so Gigerenzer provides examples.
In 2009, US Airways Flight 1549 experienced engine problems just moments after take-off from LaGuardia Airport in New York City. At that point, Captain Chesley Sullenberger, the pilot, made a quick decision to land the plane on the Hudson River. Even with some four decades of flying experience, it was highly unlikely that at that point in time, Sullenberger was of mind to rationalise and perform mathematical calculations of how his odds were stacked at that point. Yet, he arrived at a decision that allowed all 155 of his passengers and crew to return to safety.
Another common example is seen in sports, where players rely a lot on strong feelings that are not easily understood or rationalised. Citing Richard Dawkins' 'The Selfish Gene', he read, "When a man throws a ball high in the air and catches it again... he behaves as if he had solved a set of differential equations in predicting the trajectory of the ball... At some subconscious level, something functionally equivalent to the mathematical calculations is going on."
Here, Gigerenzer drew the audience's attention to the fact that Dawkins had used the words 'as if' to suggest that it cannot be true that people's brains are making mathematical calculations at such instances. Research has, however, shown that players rely on heuristics - and in this case, the gaze heuristic.
"The first is to fixate the ball with your eyes. Second, start running. Third, adjust the running speed so that the angle of the gaze remains constant." Also, the player runs in a way such that his or her gaze remains constant, ignoring variables that are irrelevant, for instance, noise, and focusing on factors that could matter, such as wind speed, wind direction, spin, etc.
How do experienced players know how to arrive at the best decisions? They have a little adaptive toolbox, said Gigerenzer. "If you ever interview soccer players; what did they do, how do they do it so well, they will usually tell you they have no idea; it's intuitive... a strong feeling about what to do but you can't explain."
A number of experiments have shown that players, like everyone here, have a little adaptive toolbox, he noted. And what this example illustrates is that complex problems do not necessarily require complex solutions. "But, for some reason, we don't think about simplicity; we always want something complex, maybe to impress someone else."
Less is more
People often think that heuristics come at the cost of accuracy and effort; that it cannot be accurate because less effort had been invested on information and computation. This is not true as "there are situations where one attains higher accuracy with less effort," Gigerenzer wrote in 'Homo Heuristicus: Why Biased Minds Make Better Inferences', a paper on which the lecture was based.
He cited a study conducted by Wübben & Wangenheim (2008) where a big company, in deciding to send a catalogue to its database of customers, used, in one condition, a heuristic that took into account if the customer had bought something in the last nine months (and everything else was ignored); and in the other, a more complex computation to predict which customers were more active / inactive.
In the latter condition, sophisticated statistical methods, such as the Pareto Negative Binomial Distribution model, took more information into account. The results showed, however, that more information did not lead to better inferences. The complex model turned out to be less accurate at predicting inactive customers, compared to the heuristic.
It is a common misconception that optimisation is better than heuristics at decision making, Gigerenzer said. "We know this is not the case because optimisation models are optimal within their mathematical assumptions. Whether they are optimal in the real world, we do not know. If just one assumption is not met, then we won't know where it's going."
Other "textbook" misconceptions: Heuristics are only useful because of human cognitive limitations; and that the more information, computation and time, the better the decision. But one only has to think about language and sports to debunk those arguments. "Studies on language acquisition indicate that there are sensitive phases in which a reduced memory and simpler input (e.g. baby talk) speeds up language acquisition," wrote Gigerenzer.
"Experiments with experienced handball players indicate that they make better decisions with less time; and expert golfers (but not novices) do better when they have only 3 seconds to putt than when the can take all the time they want... There is more to heuristics than the accuracy-effort trade-off: The mind can use less information and computation or take less time and nevertheless achieve better performance."
Herbert Simon, according to Gigerenzer, was the academic that first inspired research into heuristics. Simon proposed that instead of maximising returns, people were "satisficing" - an old English word for 'satisfying' that means arriving at a "good enough" outcome.
Gigerenzer explained the concept of "bounded rationality" by means of "Simon's Scissors"; the idea that the human mind matches the environment like two blades of a pair of scissors, complementing one another but not necessary mirroring one another. In that sense, he said people rely on heuristics based on their assessments of "rationality".
On one end, there is the "as if" model, as illustrated earlier with the ball game example. Here, people may add parameters to the 'equation' if they feel that the standard optimisation or utility maximisation theories cannot apply to the phenomenon. On the other end, there is the "homo heuristic", where rationality takes an ecological framework, influenced by the structure of the environment.
One example is the 'recognition heuristic', where people make decisions based on what they recognise. "Assume you're in the TV programme, 'Who wants to be a Millionaire'. You've made it all the way to the top. Now, you have to answer your million dollar question: Which American city has more inhabitants: Milwaukee or Detroit?" he asked the audience. A majority picked Detroit - the correct answer.
A similar test had been conducted in America and in Germany. In America, the proportion of correct responses was about 60%. In Germany, about 90% got the right answer. So how is it possible that Germans could make better inferences about American than the Americans? Gigerenzer explained that few Germans are familiar with either of the states. However, most of them have heard of Detroit while almost none had ever heard of Milwaukee.
"They used the recognition heuristic, which is that if you've never heard of Milwaukee but you've heard of Detroit, Detroit is probably larger. This isecologically rational in a world where the media often talks more about the bigger things than smaller things."
"All inductive processes, including heuristics, make bets... its accuracy is always relative to the structure of the environment," Gigerenzer wrote in his paper. The emphasis should thus be on which heuristic works better in which environment, and why a heuristic might fail or succeed.
To provide an example with practical organisational applications, Gigerenzer spoke of a study he had conducted at a hospital in Michigan. The hospital was struggling with overcrowding within their coronary care unit - close to 90% of all patients passing through the emergency room were being sent there. The reason: defensive decision making.
If a patient complains of chest pains and is rushed to an emergency room, doctors have the option of either placing the patient on a regular bed or at the coronary care unit. It could be a life-or-death decision. "He could get a heart attack at either of those places. If he doesn't get a heart attack, he wouldn't want to be at the coronary care unit. That's a dangerous place to be; people can pick up illnesses there," he said.
Still, doctors protect themselves by putting the patient into a place that lowers their potential risk. "If the patient dies, the family might sue you if the patient dies on a regular bed from a heart attack. But the doctor or the hospital will unlikely be sued if the patient dies in the coronary care unit. This is a very important concept - protection from the potential cost of failure."
Some 93% of US doctors are said to practice defensive medicine; suggesting more treatments and diagnostics than necessary, Gigerenzer said - due in part to the country's highly litigious environment. When the Michigan hospital called researchers in to help solve the overcrowding issue, the researchers took a "complex problem, complex solution" approach. The result: a prediction model based on multiple variables and logistic regression.
At the same time, a simple heuristic termed "fast and frugal tree" was constructed, based on a smaller set of criteria that asks decision makers a series of three questions. From these questions, doctors could then choose where to place the patient. "How many of you would still rather trust defensive decision making (the doctor default choice)? Or would you trust the logistic regression or the heuristic?" Gigerenzer asked the audience, before revealing the results of the study (see Figure 1)
While the statistical model showed mixed results, good and bad, it did not allow for much flexibility as the criteria and assumptions were all clearly defined. The heuristic, on the other hand, produced a positive outcome and gave decision makers greater leeway to adapt to new information. Finally, defensive decision-making, without help from models or heuristics, stood a great chance of error.
Getting rid of error
When it comes to predicting outcomes, complex models are likelier to contain error. Here, Gigerenzer said that while 'error' is a sum of 'bias', 'variance' and 'noise', people often assume, in conventional computations, that 'bias' is somehow worse than 'variance'.
Total error = (bias)2 + variance + noise
Gigerenzer conducted an analysis of London's temperature in 2000. The average temperature reading of each day was computed and patterns were established using various degrees of polynomials - the higher the degree, the better the 'fit'. And as seen from the graph below, 'degree 12 polynomial' offered a closer fit to the readings, compared to 'degree 3 polynomial'.
When analysed for error, the prediction curve (as denoted by the dotted line below, in Figure 3) shows that error goes up as the degree of polynomial increases. What this demonstrates is that while complex models may offer a bigger 'fit', they also give rise to greater error. So while too much simplicity is not optimal for decision making, too much complexity can lead to a completely unrealistic model, said Gigerenzer.
When the error was decomposed to uncover levels of 'bias' and 'variance', Gigerenzer found that as models become more complex, 'bias' falls whereas 'variance' increases (see Figure 4). "This example illustrates a fundamental problem in statistical inference known as the bias-variance dilemma. To achieve low prediction error on a broad class of problems, a model must accommodate a rich class of patterns in order to ensure low bias... Diversity in the class of patterns that the model can accommodate is, however, likely to come at a price. The price is an increase in variance," he explained in his paper.
In reducing the 'bias' from the total error, greater "effort" is required to derive a more complex polynomial. Yet, such efforts may be counterproductive as it increases the 'variance' component of error. It is perhaps paradoxical to note that more error is created as more resources are expended to guard against uncertainty. Simple heuristics may thus be better and more accurate at predicting outcomes - due not only to its flexibility, but also because it allows bias.
In an uncertain world, less is more, and that little bit of bias will allow for better inferences, he concluded.
Most Popular Slideshows
- 'Game of Thrones'-like Film, 'The Queen of the Tearling,' Casts Emma Watson as Lead Star and Exec Producer [PHOTOS]
- Kim Kardashian Baby Girl: Suggested Ways Kanye West’s New Born Can Earn Money to Keep Up with the Kardashians [PHOTOS]
- Asus Transformer Infinity Pad, Sony Vaio Duo, Toshiba Satellite, A Look at Intel's Haswell 4th Generation Ultrabooks and Notebooks [Photos]
- 'Fifty Shades of Grey' Movie Casting: Selena Gomez Open to Nude Role, Ready for Anastasia Steele? [PHOTOS]