Heuristics | Ecological Rationality | The Adaptive Toolbox | Four Classes of Heuristics
Accuracy-Effort Trade-Off
Image Credit: Wikimedia Commons
Image Credit: Wikimedia Commons

Heuristics have been and usually still are considered the most error-prone and intuitive method of decision-making. This idea is termed as the accuracy-effort trade-off: where humans use heuristics to save effort and compromise on accuracy. In contrast, the other two methods of decision-making, logic and statistics, are associated with rational reasoning. However, as the study of heuristics continues to develop, new evidence has led prominent psychologists to believe that the accuracy-effort trade-off view of heuristics is incorrect. The justification for this refutation of the accuracy-effort trade-off, simply put, points to the fact that logical and statistical methods of decision making require a ‘perfect’ world where all the necessary information is available and accurate. Since we do not live in a perfect world and information often has to be extrapolated from samples or estimated, logical and statistical models can not perfectly predict patterns in the future that would influence our decision. These methods are often referred to as ‘optimizing strategies’ or ‘models of rationality.’ In contrast, because heuristics do not require much information, they do not need a ‘perfect’ world in order to be functional and accurate. See the TED Talk video at the bottom of the page.
Image Credit: Steve Hodgson
Image Credit: Steve Hodgson

Here is one example where heuristics have proven to be more accurate than the models of rationality.

Stores need to decide which customers are more likely to shop at the store within a given time frame and classify their customers as either inactive or active based on that data. The fanciest modern approach to deciding uses the Pareto/NBD model, which uses negative binomial distribution, a statistics method, to determine which customers will be active or inactive. The heuristic alternative to the model is called the hiatus heuristic. This heuristic simply reasons that if a customer hasn’t bought anything in a certain number of months, he/she is inactive and vice versa. How does the hiatus heuristic compare to the Pareto/NBD model? According to a study done by Wubben & Wangenheim in 2008, using the hiatus heuristic correctly classified 83% of customers as inactive or active, whereas the Pareto/NBD Model only correctly classified 75% of customers. Here are the results of the studies in chart form.
Image Credit: Dave McLean
Image Credit: Dave McLean

hiatus heuristic
Pareto/NBD Model
Clothing retailer
83% accurate (83% of customers were correctly classified)
Online CD Store
Note that while the hiatus heuristic works better for classifying customers for the clothing retailer and airline, the two prediction methods tie when it comes to classifying customers for the online CD store.
(Gigerenzer, Gerd, & Gaissmaier, Wolfgang 2011).

See the table below for other incorrect assumptions about heuristics.

Six Common but Erroneous Beliefs About Heuristics

Six Common Misconceptions
1. Heuristics produce second-best results; optimization is always better.
In many situations, optimization is impossible (e.g., computationally intractable) or less accurate because of estimation errors (i.e., less robust; see investment example).
2. Our minds rely on heuristics only because of our cognitive limitations.
Characteristics of the environment (e.g., computational intractability) and of the mind make us rely on heuristics.
3. People rely on heuristics only in routine decisions of little importance.
People rely on heuristics for decisions of both low and high importance. See investment and organ donation examples.
4. People with higher cognitive capacities employ complex weighting and integration of information; those with lesser capacities use simple heuristics (related to Misconception 1).
Not supported by experimental evidence.... Cognitive capacities
seem to be linked to the adaptive selection of heuristics and seem less linked to the execution of a heuristic. See also the Markowitz example.
5. Affect, availability, causality, and representativeness are models of heuristics.
These terms are mere labels, not formal models of heuristics. A model makes precise predictions and can be tested, such as in computer simulations.
6. More information and computation is always better.
Good decisions in a partly uncertain world require ignoring part of the available information (e.g., to foster robustness). See the investment example.
(Gigerenzer, Gerd 2008).
For more information about the examples mentioned in the table, click here.

TED Talk - "Simple Heuristics Make Us Smart" - Gerd Gigerenzer

Next section: Ecological Rationality