Take the Best
Take the Best is a fast and frugal heuristic that searches for cues in order of their validity and makes a decision based on the first discriminating cue, rather than integrating all information. It is similar to Minimalist but uses the highest-validity cue first.
THE IDEA
Take the Best
This heuristic is exactly like Minimalist except that the cue with the highest validity, rather than a random cue, is tried first. If this cue does not discriminate, the next best cue is tried, and so forth. Thus, Take the Best differs from Minimalist only in Step 1.
Step 1. Ordered search: Select the cue with the highest validity and look up the cue values of the two objects.
The validity v_i of cue i is R_i / (R_i + W_i), where R_i and W_i count correct and incorrect inferences based on that cue alone. We count which inferences are right and wrong across all possible inferences in a reference class of objects.
In the source’s city dataset, Los Angeles has a cue value of 1 for rent control whereas Chicago has a cue value of 0. The rent control cue suggests that Los Angeles has a higher homelessness rate; because this matches the dataset, it counts as a right inference. Between Chicago and New York, the rent control cue makes a wrong inference. And between Chicago and New Orleans, it does not discriminate – and cannot make an inference – because both cities have 0 cue values for rent control. If we count the number of right and wrong inferences for all possible pairings of the 50 U.S. cities, we find that 90% of the inferences based on rent control are right; thus the cue validity of rent control is .90.
For instance, when inferring whether Chicago or New Orleans has a higher homelessness rate, Take the Best looks up first the cue values of the two cities for rent control because it is the cue with the highest validity (.90). Unfortunately, this cue does not discriminate – both cities have cue values of zero. So, Take the Best looks up the second-best cue, the vacancy rate cue (validity .73). This cue does discriminate, so search is stopped. Take the Best infers that Chicago, the city with the unitary cue value in contrast to New Orleans’ 0, has the higher homelessness rate.
Take the Best and Minimalist are constructed from several building blocks of fast and frugal heuristics (Gigerenzer & Goldstein, 1996). These building blocks help us both in understanding the heuristics and in generating new heuristics.
- Step-by-step procedures - a cognitive strategy that searches for some information and checks whether this is sufficient to make a decision; if not, it searches for more information, checks whether this new information is sufficient, and so on.
- Simple stopping rules - specify computationally simple conditions for halting the gathering of more cue information. Take the Best and Minimalist stop gathering further cue information if one object has a unitary (1) value for a cue and the other does not.
- One-reason decision making - once the search is stopped, the decision is based on this recent information only, the last cue considered. Trade-offs between cues never surface. The process is non-compensatory.
Some of these building blocks appear in other heuristics that are related to Take the Best. Lexicographic strategies are very close to Take the Best, but not Minimalist. Take the Best can exploit a lack of knowledge by means of the recognition heuristic. A more distantly related strategy is Elimination by Aspects (EBA), which differs in cue selection (probabilistic, not by validity) and deals with preference rather than inference. Another related strategy is classification and regression trees (CART), which involves heavy computation and optimizing.
Studies have shown that Take the Best can achieve similar accuracy to standard linear models like multiple regression while using less information and computation.
When to use it
Comparing two cases with a validated decision rule: inspect cues in order of predictive validity and stop at the first cue that distinguishes the cases.
Screening pairs of applications with cues ranked by their demonstrated ability to predict the relevant performance measure.
Comparing products on one feature at a time uses a related lexicographic strategy. Ranking features by personal importance differs from ranking cues by predictive validity.
Testing whether a coach’s ranking of predictive statistics can support player comparisons with fewer measurements.
FROM THEORY TO EVERYDAY LIFE
In practice
City Homelessness Rates
When comparing Chicago and New Orleans to decide which has a higher homelessness rate, Take the Best first checks the most valid cue (rent control, validity .90). Since both cities have 0 for rent control, it proceeds to the next best cue (vacancy rate, validity .73). Chicago has a positive value for vacancy rate, so the heuristic stops and infers Chicago has a higher homelessness rate.
City Population Inference
In the study by Gigerenzer and Goldstein (1996), Take the Best was used to infer which of two cities has a larger population. It used cues like whether the city has a major league sports team, a university, etc., ordered by their validity, and achieved accuracy comparable to multiple regression.
A MOMENT TO REFLECT
Put it to work.
In what real-world situations might ordering cues by their validity lead to better outcomes than using all available information?
How does the Take the Best heuristic balance speed and accuracy, and when might it fail?
Can you think of a decision you made recently where you intuitively used a similar one-reason strategy? How did it compare to a more thorough analysis?
Often confused with
References
- Group report: How to improve Bayesian reasoning without instruction: Frequency formats — Gigerenzer & Goldstein (1996)
- Plans and the structure of behavior — Miller, Galanter, & Pribram (1960)
- The adaptive decision maker — Payne, Bettman, & Johnson (1993)
- Contingent decision behavior — Payne, Bettman, & Johnson (1988)
- Decisions with multiple objectives: Preferences and value tradeoffs — Keeney & Raiffa (1993)
- Elimination by aspects: A theory of choice — Tversky (1972)
CONTINUE THE CONVERSATION
Source texts
- Heuristics and Biases The Psychology of Intuitive Judgment (Thomas Gilovich, Dale Griffin, Daniel Kahneman)