Take-the-best

Take-the-best is a heuristic decision strategy developed by Gerd Gigerenzer and the ABC Research Group as part of the ecological rationality program. It chooses between two alternatives by searching through available cues in order of their validity (predictive accuracy), and selecting the alternative with the first cue that discriminates between them. The heuristic ignores all other cues — it does not integrate information, weight cues, or compute any summary statistic.\n\nThe procedure is deceptively simple. Given a choice between two cities, for instance, the heuristic searches through cues like population, presence of a university, or major sports team. If the first cue (say, population) discriminates — one city is known to be larger — it stops and chooses that city. If the first cue does not discriminate, it moves to the second. The key condition for its success is that the cue validities are noncompensatory: the best cue must be substantially more predictive than any combination of weaker cues, so that no amount of weaker evidence could override it.\n\nIn environments with this structure, take-the-best matches or exceeds the performance of multiple regression, decision trees, and more complex machine learning models — despite using only a fraction of the information. The reason is not that the heuristic approximates the optimal model. It is that the heuristic exploits a structural property of the environment that complex models do not: the noncompensatory cue hierarchy.\n\nThe broader implication is that ignoring information can be rational. The "less-is-more" effect — where a heuristic with less information outperforms a model with more — occurs precisely when the additional information introduces noise that overwhelms the signal in the best cues. This inverts the standard assumption that more information is always better, and it relocates the analysis of rationality from the properties of the agent to the fit between agent and environment.\n\n\n\n