Simple Heuristics at Work in the World

imitationThis post is based on Chapter 10 of Bounded Rationality and on Gut Feelings both authored primarily by Gerd Gigerenzer. It is largely a shopping list of heuristics or rules of thumb. Gigerenzer makes the important point that more information and more choice are not always better. Less is more under certain conditions:

  1. A beneficial degree of ignorance–This seems more like a coincidence to me, but Gigerenzer includes it.
  2. Unconscious motor skills–Over deliberation can mess up skills.
  3. Cognitive limitations–Our brains can benefit by doing such things as forgetting so we do not have too much information.-Gary Marcus in Kluge would not agree or at least would not be so proud of our forgetting.
  4. Freedom of choice paradox–At some point more options create conflicts that make it more difficult to compare options. Betsch and Glockner point out as I have discussed in Intuition in J/DM that even intuition is slowed by conflicts in the data.
  5. The benefits of simplicity–In a complex world, simple rules can be better than complex rules.
  6. Information costs–Extracting too much information can hurt trust.

When those conditions exist, the following heuristics (This is not a comprehensive list.) can improve performance:

1. Imitation often works.  However, if the environment is changing rapidly, imitation will likely fail.  One rule, “imitate if better” has one imitating all others who are more successful.  In risky environments, this can lead everyone to choose the alternative with the lowest expected payoff.

2. Equal Weighting can work better than multiple regression especially as the number of independent variables increases.  Complex models like multiple regression consider all the information and find optimal weights.  However, this makes them subject to reversing the size of weights based on sampling fluctuations.  Equal weighting will also tend to be superior when the overall explanatory power of the model as measured by the r squared is relatively low.  In other words, when the model is not very strong, the unequal weights are over precise.

3. Take the Best is a lexicographic procedure that uses a rank ordering of cues. Cues are searched through one at a time until a cue that satisfies a stopping rule is found.  The decision is then made solely on the cue that stopped the search.  Take the best uses only about one third of information available, but it can work because it focuses on the highest ranked cue. This heuristic includes:

  1. A search rule–look up reasons in the order of importance.
  2. A stopping rule- stop search as soon as alternatives differ.
  3. A decision rule- choose alternative that the reason suggests.

Gigerenzer points out that the Arabic number system is lexicographic as opposed to the Roman numbering system.  MDCCCLXXX looks like a bigger number than MCMXI, but it is not as is readily notable with equal Arabic numbers 1911 and 1880.  In the Arabic system, if two numbers have the same length, one only need to search from left to right for the first digit that is different.  One can then stop the search and conclude that the number with the higher digit is the larger number.


4. Take the First is a recognition primed strategy for experts.  It has been studied in pilots, firefighters, and chess players, and as it states the best course of action may be to take the first or only solution that comes to mind. It is argued to be effective because, for an expert, part of categorizing a situation as typical is to recall what to do in that situation and options are not generated randomly, but possibly in order of quality.  Obviously, this strategy is not useful in novel situations or areas where you are not an expert.

5. Small sample inferences are more or less jumping to conclusions.  With our limited working memories, small sample of experience, and ability to detect covariances, we can benefit in tasks and domains where the costs of missing a relationship are high. Ironically, in such situations, people with smaller working memories perform better.

6. The Recognition heuristic is based on the idea that more important things tend to be recognized more because they are talked about or read about more.  Thus, you tend to hear about bigger cities more so if you recognize only one of two cities, it is likely that the recognized city is larger in population.  The recognition heuristic is based on missing knowledge on a systematic basis. People who use the recognition heuristic can make snap decisions which sees to impress those who knew more, but needed time to reflect.

Of course, with heuristics, we have to decide how to decide which to use. There can be the top down strategy which is consistent with people planning how to solve problems. There is the bottom up approach which reflects experience with the various heuristics, but which does not explain what to use in a novel situation.  A more opportunistic approach may also be used where we start off and then switch heuristics. Gigerenzer adds a mystical touch:  “The intelligence of the unconscious is in knowing, without thinking, which rule is likely to work in which situation.”

Gigerenzer, G.(2007) Gut Feelings: the Intelligence of the Unconscious. New York:  Viking.

Gigerenzer, G.& Selten R. editors (1999) Bounded Rationality the Adaptive Toolbox. Cambridge, MA:  MIT Press.